Demo uploaded with RVC options and more languages

app.py

@@ -1,7 +1,21 @@
 #%cd SoniTranslate
+# vc infer pipe 161 np.int
+import os
+os.system("pip install -r requirements_colab.txt")
+os.system("pip install -r requirements_extra.txt")
+
+os.system('apt install git-lfs')
+os.system('git lfs install')
+os.system('apt -y install -qq aria2')
+os.system('aria2c --console-log-level=error -c -x 16 -s 16 -k 1M https://huggingface.co/lj1995/VoiceConversionWebUI/resolve/main/hubert_base.pt -d . -o hubert_base.pt')
+os.system('wget https://huggingface.co/lj1995/VoiceConversionWebUI/resolve/main/rmvpe.pt')
+
 import numpy as np
 import gradio as gr
 import whisperx
+from whisperx.utils import LANGUAGES as LANG_TRANSCRIPT
+from whisperx.alignment import DEFAULT_ALIGN_MODELS_TORCH as DAMT, DEFAULT_ALIGN_MODELS_HF as DAMHF
+from IPython.utils import capture
 import torch
 from gtts import gTTS
 import librosa
@@ -15,14 +29,29 @@ import os
 from soni_translate.audio_segments import create_translated_audio
 from soni_translate.text_to_speech import make_voice_gradio
 from soni_translate.translate_segments import translate_text
+import time
+import shutil
+from urllib.parse import unquote
+import zipfile
+import rarfile
+
+
 
 title = "<center><strong><font size='7'>📽️ SoniTranslate 🈷️</font></strong></center>"
 
 news = """ ## 📖 News
-        🔥 2023/07/26: new UI and mix options add.
+        🔥 2023/07/26: New UI and add mix options.
+
+        🔥 2023/07/27: Fix some bug processing the video and audio.
+
+        🔥 2023/08/01: Add options for use RVC models.
+
+        🔥 2023/08/02: Added support for Arabic, Czech, Danish, Finnish, Greek, Hebrew, Hungarian, Korean, Persian, Polish, Russian, Turkish, Urdu, Hindi, and Vietnamese languages. 🌐
+
+        🔥 2023/08/03: Changed default options and added directory view of downloads..
         """
 
-description = """ 
+description = """
 ### 🎥 **Translate videos easily with SoniTranslate!** 📽️
 
 Upload a video or provide a video link. Limitation: 10 seconds for CPU, but no restrictions with a GPU.
@@ -32,13 +61,14 @@ For faster results and no duration limits, try the Colab notebook with a GPU:
 
 📽️ **This a demo of SoniTranslate; GitHub repository: [SoniTranslate](https://github.com/R3gm/SoniTranslate)!**
 
-See the tab labeled 'Help' for instructions on how to use it. Let's start having fun with video translation! 🚀🎉
+See the tab labeled `Help` for instructions on how to use it. Let's start having fun with video translation! 🚀🎉
 """
 
 
 
-tutorial = """ 
-## 🔰 **Instructions for use:**
+tutorial = """
+
+# 🔰 **Instructions for use:**
 
 1. 📤 **Upload a video** on the first tab or 🌐 **use a video link** on the second tab.
 
@@ -47,6 +77,24 @@ tutorial = """
 3. 🗣️ Specify the **number of people speaking** in the video and **assign each one a text-to-speech voice** suitable for the translation language.
 
 4. 🚀 Press the '**Translate**' button to obtain the results.
+
+
+# 🎤 How to Use RVC and RVC2 Voices 🎶
+
+The goal is to apply a RVC (Retrieval-based Voice Conversion) to the generated TTS (Text-to-Speech) 🎙️
+
+1. In the `Custom Voice RVC` tab, download the models you need 📥 You can use links from Hugging Face and Google Drive in formats like zip, pth, or index. You can also download complete HF space repositories, but this option is not very stable 😕
+
+2. Now, go to `Replace voice: TTS to RVC` and check the `enable` box ✅ After this, you can choose the models you want to apply to each TTS speaker 👩‍🦰👨‍🦱👩‍🦳👨‍🦲
+
+3. Adjust the F0 method that will be applied to all RVCs 🎛️
+
+4. Press `APPLY CONFIGURATION` to apply the changes you made 🔄
+
+5. Go back to the video translation tab and click on 'Translate' ▶️ Now, the translation will be done applying the RVCs 🗣️
+
+Tip: You can use `Test RVC` to experiment and find the best TTS or configurations to apply to the RVC 🧪🔍
+
 """
 
 
@@ -56,16 +104,171 @@ if torch.cuda.is_available():
     device = "cuda"
     list_compute_type = ['float16', 'float32']
     compute_type_default = 'float16'
-    whisper_model_default = 'large-v1'
+    whisper_model_default = 'large-v2'
 else:
     device = "cpu"
     list_compute_type = ['float32']
     compute_type_default = 'float32'
-    whisper_model_default = 'base'
+    whisper_model_default = 'medium'
 print('Working in: ', device)
 
 list_tts = ['af-ZA-AdriNeural-Female', 'af-ZA-WillemNeural-Male', 'am-ET-AmehaNeural-Male', 'am-ET-MekdesNeural-Female', 'ar-AE-FatimaNeural-Female', 'ar-AE-HamdanNeural-Male', 'ar-BH-AliNeural-Male', 'ar-BH-LailaNeural-Female', 'ar-DZ-AminaNeural-Female', 'ar-DZ-IsmaelNeural-Male', 'ar-EG-SalmaNeural-Female', 'ar-EG-ShakirNeural-Male', 'ar-IQ-BasselNeural-Male', 'ar-IQ-RanaNeural-Female', 'ar-JO-SanaNeural-Female', 'ar-JO-TaimNeural-Male', 'ar-KW-FahedNeural-Male', 'ar-KW-NouraNeural-Female', 'ar-LB-LaylaNeural-Female', 'ar-LB-RamiNeural-Male', 'ar-LY-ImanNeural-Female', 'ar-LY-OmarNeural-Male', 'ar-MA-JamalNeural-Male', 'ar-MA-MounaNeural-Female', 'ar-OM-AbdullahNeural-Male', 'ar-OM-AyshaNeural-Female', 'ar-QA-AmalNeural-Female', 'ar-QA-MoazNeural-Male', 'ar-SA-HamedNeural-Male', 'ar-SA-ZariyahNeural-Female', 'ar-SY-AmanyNeural-Female', 'ar-SY-LaithNeural-Male', 'ar-TN-HediNeural-Male', 'ar-TN-ReemNeural-Female', 'ar-YE-MaryamNeural-Female', 'ar-YE-SalehNeural-Male', 'az-AZ-BabekNeural-Male', 'az-AZ-BanuNeural-Female', 'bg-BG-BorislavNeural-Male', 'bg-BG-KalinaNeural-Female', 'bn-BD-NabanitaNeural-Female', 'bn-BD-PradeepNeural-Male', 'bn-IN-BashkarNeural-Male', 'bn-IN-TanishaaNeural-Female', 'bs-BA-GoranNeural-Male', 'bs-BA-VesnaNeural-Female', 'ca-ES-EnricNeural-Male', 'ca-ES-JoanaNeural-Female', 'cs-CZ-AntoninNeural-Male', 'cs-CZ-VlastaNeural-Female', 'cy-GB-AledNeural-Male', 'cy-GB-NiaNeural-Female', 'da-DK-ChristelNeural-Female', 'da-DK-JeppeNeural-Male', 'de-AT-IngridNeural-Female', 'de-AT-JonasNeural-Male', 'de-CH-JanNeural-Male', 'de-CH-LeniNeural-Female', 'de-DE-AmalaNeural-Female', 'de-DE-ConradNeural-Male', 'de-DE-KatjaNeural-Female', 'de-DE-KillianNeural-Male', 'el-GR-AthinaNeural-Female', 'el-GR-NestorasNeural-Male', 'en-AU-NatashaNeural-Female', 'en-AU-WilliamNeural-Male', 'en-CA-ClaraNeural-Female', 'en-CA-LiamNeural-Male', 'en-GB-LibbyNeural-Female', 'en-GB-MaisieNeural-Female', 'en-GB-RyanNeural-Male', 'en-GB-SoniaNeural-Female', 'en-GB-ThomasNeural-Male', 'en-HK-SamNeural-Male', 'en-HK-YanNeural-Female', 'en-IE-ConnorNeural-Male', 'en-IE-EmilyNeural-Female', 'en-IN-NeerjaExpressiveNeural-Female', 'en-IN-NeerjaNeural-Female', 'en-IN-PrabhatNeural-Male', 'en-KE-AsiliaNeural-Female', 'en-KE-ChilembaNeural-Male', 'en-NG-AbeoNeural-Male', 'en-NG-EzinneNeural-Female', 'en-NZ-MitchellNeural-Male', 'en-NZ-MollyNeural-Female', 'en-PH-JamesNeural-Male', 'en-PH-RosaNeural-Female', 'en-SG-LunaNeural-Female', 'en-SG-WayneNeural-Male', 'en-TZ-ElimuNeural-Male', 'en-TZ-ImaniNeural-Female', 'en-US-AnaNeural-Female', 'en-US-AriaNeural-Female', 'en-US-ChristopherNeural-Male', 'en-US-EricNeural-Male', 'en-US-GuyNeural-Male', 'en-US-JennyNeural-Female', 'en-US-MichelleNeural-Female', 'en-US-RogerNeural-Male', 'en-US-SteffanNeural-Male', 'en-ZA-LeahNeural-Female', 'en-ZA-LukeNeural-Male', 'es-AR-ElenaNeural-Female', 'es-AR-TomasNeural-Male', 'es-BO-MarceloNeural-Male', 'es-BO-SofiaNeural-Female', 'es-CL-CatalinaNeural-Female', 'es-CL-LorenzoNeural-Male', 'es-CO-GonzaloNeural-Male', 'es-CO-SalomeNeural-Female', 'es-CR-JuanNeural-Male', 'es-CR-MariaNeural-Female', 'es-CU-BelkysNeural-Female', 'es-CU-ManuelNeural-Male', 'es-DO-EmilioNeural-Male', 'es-DO-RamonaNeural-Female', 'es-EC-AndreaNeural-Female', 'es-EC-LuisNeural-Male', 'es-ES-AlvaroNeural-Male', 'es-ES-ElviraNeural-Female', 'es-GQ-JavierNeural-Male', 'es-GQ-TeresaNeural-Female', 'es-GT-AndresNeural-Male', 'es-GT-MartaNeural-Female', 'es-HN-CarlosNeural-Male', 'es-HN-KarlaNeural-Female', 'es-MX-DaliaNeural-Female', 'es-MX-JorgeNeural-Male', 'es-NI-FedericoNeural-Male', 'es-NI-YolandaNeural-Female', 'es-PA-MargaritaNeural-Female', 'es-PA-RobertoNeural-Male', 'es-PE-AlexNeural-Male', 'es-PE-CamilaNeural-Female', 'es-PR-KarinaNeural-Female', 'es-PR-VictorNeural-Male', 'es-PY-MarioNeural-Male', 'es-PY-TaniaNeural-Female', 'es-SV-LorenaNeural-Female', 'es-SV-RodrigoNeural-Male', 'es-US-AlonsoNeural-Male', 'es-US-PalomaNeural-Female', 'es-UY-MateoNeural-Male', 'es-UY-ValentinaNeural-Female', 'es-VE-PaolaNeural-Female', 'es-VE-SebastianNeural-Male', 'et-EE-AnuNeural-Female', 'et-EE-KertNeural-Male', 'fa-IR-DilaraNeural-Female', 'fa-IR-FaridNeural-Male', 'fi-FI-HarriNeural-Male', 'fi-FI-NooraNeural-Female', 'fil-PH-AngeloNeural-Male', 'fil-PH-BlessicaNeural-Female', 'fr-BE-CharlineNeural-Female', 'fr-BE-GerardNeural-Male', 'fr-CA-AntoineNeural-Male', 'fr-CA-JeanNeural-Male', 'fr-CA-SylvieNeural-Female', 'fr-CH-ArianeNeural-Female', 'fr-CH-FabriceNeural-Male', 'fr-FR-DeniseNeural-Female', 'fr-FR-EloiseNeural-Female', 'fr-FR-HenriNeural-Male', 'ga-IE-ColmNeural-Male', 'ga-IE-OrlaNeural-Female', 'gl-ES-RoiNeural-Male', 'gl-ES-SabelaNeural-Female', 'gu-IN-DhwaniNeural-Female', 'gu-IN-NiranjanNeural-Male', 'he-IL-AvriNeural-Male', 'he-IL-HilaNeural-Female', 'hi-IN-MadhurNeural-Male', 'hi-IN-SwaraNeural-Female', 'hr-HR-GabrijelaNeural-Female', 'hr-HR-SreckoNeural-Male', 'hu-HU-NoemiNeural-Female', 'hu-HU-TamasNeural-Male', 'id-ID-ArdiNeural-Male', 'id-ID-GadisNeural-Female', 'is-IS-GudrunNeural-Female', 'is-IS-GunnarNeural-Male', 'it-IT-DiegoNeural-Male', 'it-IT-ElsaNeural-Female', 'it-IT-IsabellaNeural-Female', 'ja-JP-KeitaNeural-Male', 'ja-JP-NanamiNeural-Female', 'jv-ID-DimasNeural-Male', 'jv-ID-SitiNeural-Female', 'ka-GE-EkaNeural-Female', 'ka-GE-GiorgiNeural-Male', 'kk-KZ-AigulNeural-Female', 'kk-KZ-DauletNeural-Male', 'km-KH-PisethNeural-Male', 'km-KH-SreymomNeural-Female', 'kn-IN-GaganNeural-Male', 'kn-IN-SapnaNeural-Female', 'ko-KR-InJoonNeural-Male', 'ko-KR-SunHiNeural-Female', 'lo-LA-ChanthavongNeural-Male', 'lo-LA-KeomanyNeural-Female', 'lt-LT-LeonasNeural-Male', 'lt-LT-OnaNeural-Female', 'lv-LV-EveritaNeural-Female', 'lv-LV-NilsNeural-Male', 'mk-MK-AleksandarNeural-Male', 'mk-MK-MarijaNeural-Female', 'ml-IN-MidhunNeural-Male', 'ml-IN-SobhanaNeural-Female', 'mn-MN-BataaNeural-Male', 'mn-MN-YesuiNeural-Female', 'mr-IN-AarohiNeural-Female', 'mr-IN-ManoharNeural-Male', 'ms-MY-OsmanNeural-Male', 'ms-MY-YasminNeural-Female', 'mt-MT-GraceNeural-Female', 'mt-MT-JosephNeural-Male', 'my-MM-NilarNeural-Female', 'my-MM-ThihaNeural-Male', 'nb-NO-FinnNeural-Male', 'nb-NO-PernilleNeural-Female', 'ne-NP-HemkalaNeural-Female', 'ne-NP-SagarNeural-Male', 'nl-BE-ArnaudNeural-Male', 'nl-BE-DenaNeural-Female', 'nl-NL-ColetteNeural-Female', 'nl-NL-FennaNeural-Female', 'nl-NL-MaartenNeural-Male', 'pl-PL-MarekNeural-Male', 'pl-PL-ZofiaNeural-Female', 'ps-AF-GulNawazNeural-Male', 'ps-AF-LatifaNeural-Female', 'pt-BR-AntonioNeural-Male', 'pt-BR-FranciscaNeural-Female', 'pt-PT-DuarteNeural-Male', 'pt-PT-RaquelNeural-Female', 'ro-RO-AlinaNeural-Female', 'ro-RO-EmilNeural-Male', 'ru-RU-DmitryNeural-Male', 'ru-RU-SvetlanaNeural-Female', 'si-LK-SameeraNeural-Male', 'si-LK-ThiliniNeural-Female', 'sk-SK-LukasNeural-Male', 'sk-SK-ViktoriaNeural-Female', 'sl-SI-PetraNeural-Female', 'sl-SI-RokNeural-Male', 'so-SO-MuuseNeural-Male', 'so-SO-UbaxNeural-Female', 'sq-AL-AnilaNeural-Female', 'sq-AL-IlirNeural-Male', 'sr-RS-NicholasNeural-Male', 'sr-RS-SophieNeural-Female', 'su-ID-JajangNeural-Male', 'su-ID-TutiNeural-Female', 'sv-SE-MattiasNeural-Male', 'sv-SE-SofieNeural-Female', 'sw-KE-RafikiNeural-Male', 'sw-KE-ZuriNeural-Female', 'sw-TZ-DaudiNeural-Male', 'sw-TZ-RehemaNeural-Female', 'ta-IN-PallaviNeural-Female', 'ta-IN-ValluvarNeural-Male', 'ta-LK-KumarNeural-Male', 'ta-LK-SaranyaNeural-Female', 'ta-MY-KaniNeural-Female', 'ta-MY-SuryaNeural-Male', 'ta-SG-AnbuNeural-Male', 'ta-SG-VenbaNeural-Female', 'te-IN-MohanNeural-Male', 'te-IN-ShrutiNeural-Female', 'th-TH-NiwatNeural-Male', 'th-TH-PremwadeeNeural-Female', 'tr-TR-AhmetNeural-Male', 'tr-TR-EmelNeural-Female', 'uk-UA-OstapNeural-Male', 'uk-UA-PolinaNeural-Female', 'ur-IN-GulNeural-Female', 'ur-IN-SalmanNeural-Male', 'ur-PK-AsadNeural-Male', 'ur-PK-UzmaNeural-Female', 'uz-UZ-MadinaNeural-Female', 'uz-UZ-SardorNeural-Male', 'vi-VN-HoaiMyNeural-Female', 'vi-VN-NamMinhNeural-Male', 'zh-CN-XiaoxiaoNeural-Female', 'zh-CN-XiaoyiNeural-Female', 'zh-CN-YunjianNeural-Male', 'zh-CN-YunxiNeural-Male', 'zh-CN-YunxiaNeural-Male', 'zh-CN-YunyangNeural-Male', 'zh-CN-liaoning-XiaobeiNeural-Female', 'zh-CN-shaanxi-XiaoniNeural-Female']
 
+### voices
+with capture.capture_output() as cap:
+    os.system('mkdir downloads')
+    os.system('mkdir logs')
+    os.system('mkdir weights')
+    os.system('mkdir downloads')
+    del cap
+
+
+def print_tree_directory(root_dir, indent=''):
+    if not os.path.exists(root_dir):
+        print(f"{indent}Invalid directory or file: {root_dir}")
+        return
+
+    items = os.listdir(root_dir)
+
+    for index, item in enumerate(sorted(items)):
+        item_path = os.path.join(root_dir, item)
+        is_last_item = index == len(items) - 1
+
+        if os.path.isfile(item_path) and item_path.endswith('.zip'):
+            with zipfile.ZipFile(item_path, 'r') as zip_file:
+                print(f"{indent}{'└──' if is_last_item else '├──'} {item} (zip file)")
+                zip_contents = zip_file.namelist()
+                for zip_item in sorted(zip_contents):
+                    print(f"{indent}{'    ' if is_last_item else '│   '}{zip_item}")
+        else:
+            print(f"{indent}{'└──' if is_last_item else '├──'} {item}")
+
+            if os.path.isdir(item_path):
+                new_indent = indent + ('    ' if is_last_item else '│   ')
+                print_tree_directory(item_path, new_indent)
+
+
+def upload_model_list():
+    weight_root = "weights"
+    models = []
+    for name in os.listdir(weight_root):
+        if name.endswith(".pth"):
+            models.append(name)
+
+    index_root = "logs"
+    index_paths = []
+    for name in os.listdir(index_root):
+        if name.endswith(".index"):
+            index_paths.append("logs/"+name)
+
+    print(models, index_paths)
+    return models, index_paths
+
+def manual_download(url, dst):
+    token = os.getenv("YOUR_HF_TOKEN")
+    user_header = f"\"Authorization: Bearer {token}\""
+
+    if 'drive.google' in url:
+        print("Drive link")
+        if 'folders' in url:
+            print("folder")
+            os.system(f'gdown --folder "{url}" -O {dst} --fuzzy -c')
+        else:
+            print("single")
+            os.system(f'gdown "{url}" -O {dst} --fuzzy -c')
+    elif 'huggingface' in url:
+        print("HuggingFace link")
+        if '/blob/' in url or '/resolve/' in url:
+          if '/blob/' in url:
+              url = url.replace('/blob/', '/resolve/')
+          #parsed_link = '\n{}\n\tout={}'.format(url, unquote(url.split('/')[-1]))
+          #os.system(f'echo -e "{parsed_link}" | aria2c --header={user_header} --console-log-level=error --summary-interval=10 -i- -j5 -x16 -s16 -k1M -c -d "{dst}"')
+          os.system(f"wget -P {dst} {url}")
+        else:
+          os.system(f"git clone {url} {dst+'repo/'}")
+    elif 'http' in url or 'magnet' in url:
+        parsed_link = '"{}"'.format(url)
+        os.system(f'aria2c --optimize-concurrent-downloads --console-log-level=error --summary-interval=10 -j5 -x16 -s16 -k1M -c -d {dst} -Z {parsed_link}')
+
+
+def download_list(text_downloads):
+    try:
+      urls = [elem.strip() for elem in text_downloads.split(',')]
+    except:
+      return 'No valid link'
+
+    os.system('mkdir downloads')
+    os.system('mkdir logs')
+    os.system('mkdir weights')
+    path_download = "downloads/"
+    for url in urls:
+      manual_download(url, path_download)
+    
+    # Tree
+    print('####################################')
+    print_tree_directory("downloads", indent='')
+    print('####################################')
+
+    # Place files
+    select_zip_and_rar_files("downloads/")
+
+    models, _ = upload_model_list()
+    os.system("rm -rf downloads/repo")
+
+    return f"Downloaded = {models}"
+
+
+def select_zip_and_rar_files(directory_path="downloads/"):
+    #filter
+    zip_files = []
+    rar_files = []
+
+    for file_name in os.listdir(directory_path):
+        if file_name.endswith(".zip"):
+            zip_files.append(file_name)
+        elif file_name.endswith(".rar"):
+            rar_files.append(file_name)
+
+    # extract
+    for file_name in zip_files:
+        file_path = os.path.join(directory_path, file_name)
+        with zipfile.ZipFile(file_path, 'r') as zip_ref:
+            zip_ref.extractall(directory_path)
+
+    for file_name in rar_files:
+        file_path = os.path.join(directory_path, file_name)
+        with rarfile.RarFile(file_path, 'r') as rar_ref:
+            rar_ref.extractall(directory_path)
+
+    # set in path
+    def move_files_with_extension(src_dir, extension, destination_dir):
+        for root, _, files in os.walk(src_dir):
+            for file_name in files:
+                if file_name.endswith(extension):
+                    source_file = os.path.join(root, file_name)
+                    destination = os.path.join(destination_dir, file_name)
+                    shutil.move(source_file, destination)
+
+    move_files_with_extension(directory_path, ".index", "logs/")
+    move_files_with_extension(directory_path, ".pth", "weights/")
+
+    return 'Download complete'
+
+def custom_model_voice_enable(enable_custom_voice):
+    if enable_custom_voice:
+      os.environ["VOICES_MODELS"] = 'ENABLE'
+    else:
+      os.environ["VOICES_MODELS"] = 'DISABLE'
+
+
+models, index_paths = upload_model_list()
+
+f0_methods_voice = ["pm", "harvest", "crepe", "rmvpe"]
+
+
+from voice_main import ClassVoices
+voices = ClassVoices()
+
 '''
 def translate_from_video(video, WHISPER_MODEL_SIZE, batch_size, compute_type,
                          TRANSLATE_AUDIO_TO, min_speakers, max_speakers,
@@ -106,32 +309,55 @@ def translate_from_video(
     tts_voice05="en-GB-MaisieNeural-Female",
     video_output="video_dub.mp4",
     AUDIO_MIX_METHOD='Adjusting volumes and mixing audio',
+    progress=gr.Progress(),
     ):
 
     if YOUR_HF_TOKEN == "" or YOUR_HF_TOKEN == None:
       YOUR_HF_TOKEN = os.getenv("YOUR_HF_TOKEN")
       if YOUR_HF_TOKEN == None:
         print('No valid token')
-        return 
+        return "No valid token"
+      else:
+        os.environ["YOUR_HF_TOKEN"] = YOUR_HF_TOKEN
+
+    video = video if isinstance(video, str) else video.name
+    print(video)
 
     if "SET_LIMIT" == os.getenv("DEMO"):
       preview=True
       print("DEMO; set preview=True; The generation is **limited to 10 seconds** to prevent errors with the CPU. If you use a GPU, you won't have any of these limitations.")
       AUDIO_MIX_METHOD='Adjusting volumes and mixing audio'
       print("DEMO; set Adjusting volumes and mixing audio")
+      WHISPER_MODEL_SIZE="medium"
+      print("DEMO; set whisper model to medium")
 
     LANGUAGES = {
         'Automatic detection': 'Automatic detection',
+        'Arabic (ar)': 'ar',
+        'Chinese (zh)': 'zh',
+        'Czech (cs)': 'cs',
+        'Danish (da)': 'da',
+        'Dutch (nl)': 'nl',
         'English (en)': 'en',
+        'Finnish (fi)': 'fi',
         'French (fr)': 'fr',
         'German (de)': 'de',
-        'Spanish (es)': 'es',
+        'Greek (el)': 'el',
+        'Hebrew (he)': 'he',
+        'Hungarian (hu)': 'hu',
         'Italian (it)': 'it',
         'Japanese (ja)': 'ja',
-        'Chinese (zh)': 'zh',
-        'Dutch (nl)': 'nl',
+        'Korean (ko)': 'ko',
+        'Persian (fa)': 'fa',
+        'Polish (pl)': 'pl',
+        'Portuguese (pt)': 'pt',
+        'Russian (ru)': 'ru',
+        'Spanish (es)': 'es',
+        'Turkish (tr)': 'tr',
         'Ukrainian (uk)': 'uk',
-        'Portuguese (pt)': 'pt'
+        'Urdu (ur)': 'ur',
+        'Vietnamese (vi)': 'vi',
+        'Hindi (hi)': 'hi',
     }
 
     TRANSLATE_AUDIO_TO = LANGUAGES[TRANSLATE_AUDIO_TO]
@@ -157,14 +383,41 @@ def translate_from_video(
     os.system("rm audio.webm")
     os.system("rm audio.wav")
 
+    progress(0.15, desc="Processing video...")
     if os.path.exists(video):
         if preview:
             print('Creating a preview video of 10 seconds, to disable this option, go to advanced settings and turn off preview.')
             os.system(f'ffmpeg -y -i "{video}" -ss 00:00:20 -t 00:00:10 -c:v libx264 -c:a aac -strict experimental Video.mp4')
         else:
-            os.system(f'ffmpeg -y -i "{video}" -c:v libx264 -c:a aac -strict experimental Video.mp4')
+            # Check if the file ends with ".mp4" extension
+            if video.endswith(".mp4"):
+                destination_path = os.path.join(os.getcwd(), "Video.mp4")
+                shutil.copy(video, destination_path)
+            else:
+                print("File does not have the '.mp4' extension. Converting video.")
+                os.system(f'ffmpeg -y -i "{video}" -c:v libx264 -c:a aac -strict experimental Video.mp4')
+
+        for i in range (120):
+            time.sleep(1)
+            print('process video...')
+            if os.path.exists(OutputFile):
+                time.sleep(1)
+                os.system("ffmpeg -y -i Video.mp4 -vn -acodec pcm_s16le -ar 44100 -ac 2 audio.wav")
+                time.sleep(1)
+                break
+            if i == 119:
+              print('Error processing video')
+              return
+
+        for i in range (120):
+            time.sleep(1)
+            print('process audio...')
+            if os.path.exists(audio_wav):
+                break
+            if i == 119:
+              print("Error can't create the audio")
+              return
 
-        os.system("ffmpeg -y -i Video.mp4 -vn -acodec pcm_s16le -ar 44100 -ac 2 audio.wav")
     else:
         if preview:
             print('Creating a preview from the link, 10 seconds to disable this option, go to advanced settings and turn off preview.')
@@ -191,25 +444,43 @@ def translate_from_video(
                   return
 
     print("Set file complete.")
+    progress(0.30, desc="Transcribing...")
 
     SOURCE_LANGUAGE = None if SOURCE_LANGUAGE == 'Automatic detection' else SOURCE_LANGUAGE
 
     # 1. Transcribe with original whisper (batched)
-    model = whisperx.load_model(
-        WHISPER_MODEL_SIZE,
-        device,
-        compute_type=compute_type,
-        language= SOURCE_LANGUAGE,
-        )
+    with capture.capture_output() as cap:
+      model = whisperx.load_model(
+          WHISPER_MODEL_SIZE,
+          device,
+          compute_type=compute_type,
+          language= SOURCE_LANGUAGE,
+          )
+      del cap
     audio = whisperx.load_audio(audio_wav)
     result = model.transcribe(audio, batch_size=batch_size)
     gc.collect(); torch.cuda.empty_cache(); del model
     print("Transcript complete")
 
+    
+
     # 2. Align whisper output
+    progress(0.45, desc="Aligning...")
+    DAMHF.update(DAMT) #lang align
+    EXTRA_ALIGN = {
+        "hi": "theainerd/Wav2Vec2-large-xlsr-hindi"
+    } # add new align models here
+    #print(result['language'], DAM.keys(), EXTRA_ALIGN.keys())
+    if not result['language'] in DAMHF.keys() and not result['language'] in EXTRA_ALIGN.keys():
+        audio = result = None
+        print("Automatic detection: Source language not incompatible")
+        print(f"Detected language {LANG_TRANSCRIPT[result['language']]}  incompatible, you can select the source language to avoid this error.")
+        return
+
     model_a, metadata = whisperx.load_align_model(
         language_code=result["language"],
-        device=device
+        device=device,
+        model_name = None if result["language"] in DAMHF.keys() else EXTRA_ALIGN[result["language"]]
         )
     result = whisperx.align(
         result["segments"],
@@ -223,11 +494,14 @@ def translate_from_video(
     print("Align complete")
 
     if result['segments'] == []:
-      print('No active speech found in audio')
-      return
+        print('No active speech found in audio')
+        return
 
     # 3. Assign speaker labels
-    diarize_model = whisperx.DiarizationPipeline(use_auth_token=YOUR_HF_TOKEN, device=device)
+    progress(0.60, desc="Diarizing...")
+    with capture.capture_output() as cap:
+      diarize_model = whisperx.DiarizationPipeline(use_auth_token=YOUR_HF_TOKEN, device=device)
+      del cap
     diarize_segments = diarize_model(
         audio_wav,
         min_speakers=min_speakers,
@@ -236,10 +510,17 @@ def translate_from_video(
     gc.collect(); torch.cuda.empty_cache(); del diarize_model
     print("Diarize complete")
 
+    progress(0.75, desc="Translating...")
+    if TRANSLATE_AUDIO_TO == "zh":
+        TRANSLATE_AUDIO_TO = "zh-CN"
+    if TRANSLATE_AUDIO_TO == "he":
+        TRANSLATE_AUDIO_TO = "iw"
     result_diarize['segments'] = translate_text(result_diarize['segments'], TRANSLATE_AUDIO_TO)
     print("Translation complete")
 
+    progress(0.85, desc="Text_to_speech...")
     audio_files = []
+    speakers_list = []
 
     # Mapping speakers to voice variables
     speaker_to_voice = {
@@ -262,7 +543,7 @@ def translate_from_video(
         except KeyError:
             segment['speaker'] = "SPEAKER_99"
             speaker = segment['speaker']
-            print("NO SPEAKER DETECT IN SEGMENT")
+            print(f"NO SPEAKER DETECT IN SEGMENT: TTS auxiliary will be used in the segment time {segment['start'], segment['text']}")
 
         # make the tts audio
         filename = f"audio/{start}.ogg"
@@ -301,11 +582,20 @@ def translate_from_video(
 
         duration_create = librosa.get_duration(filename=f"audio2/{filename}")
         audio_files.append(filename)
+        speakers_list.append(speaker)
+
+    # custom voice
+    if os.getenv('VOICES_MODELS') == 'ENABLE':
+        progress(0.90, desc="Applying customized voices...")
+        voices(speakers_list, audio_files)
 
     # replace files with the accelerates
     os.system("mv -f audio2/audio/*.ogg audio/")
 
     os.system(f"rm {Output_name_file}")
+
+    progress(0.95, desc="Creating final translated video...")
+
     create_translated_audio(result_diarize, audio_files, Output_name_file)
 
     os.system(f"rm {mix_audio}")
@@ -320,7 +610,7 @@ def translate_from_video(
             os.system(f'ffmpeg -i {audio_wav} -i {Output_name_file} -filter_complex "[1:a]asplit=2[sc][mix];[0:a][sc]sidechaincompress=threshold=0.003:ratio=20[bg]; [bg][mix]amerge[final]" -map [final] {mix_audio}')
         except:
             # volume mix except
-            os.system(f'ffmpeg -y -i {audio_wav} -i {Output_name_file} -filter_complex "[0:0]volume=0.15[a];[1:0]volume=1.90[b];[a][b]amix=inputs=2:duration=longest" -c:a libmp3lame {mix_audio}')
+            os.system(f'ffmpeg -y -i {audio_wav} -i {Output_name_file} -filter_complex "[0:0]volume=0.25[a];[1:0]volume=1.80[b];[a][b]amix=inputs=2:duration=longest" -c:a libmp3lame {mix_audio}')
 
     os.system(f"rm {video_output}")
     os.system(f"ffmpeg -i {OutputFile} -i {mix_audio} -c:v copy -c:a copy -map 0:v -map 1:a -shortest {video_output}")
@@ -352,6 +642,10 @@ def read_logs():
     with open("output.log", "r") as f:
         return f.read()
 
+def submit_file_func(file):
+    print(file.name)
+    return file.name, file.name
+
 # max tts
 MAX_TTS = 6
 
@@ -362,12 +656,16 @@ with gr.Blocks(theme=theme) as demo:
     gr.Markdown(description)
 
 #### video
-    with gr.Tab("Translate audio from video"):
+    with gr.Tab("Audio Translation for a Video"):
         with gr.Row():
             with gr.Column():
-                video_input = gr.Video() # height=300,width=300
-                SOURCE_LANGUAGE = gr.Dropdown(['Automatic detection', 'English (en)', 'French (fr)', 'German (de)', 'Spanish (es)', 'Italian (it)', 'Japanese (ja)', 'Chinese (zh)', 'Dutch (nl)', 'Ukrainian (uk)', 'Portuguese (pt)'], value='Automatic detection',label = 'Source language', info="This is the original language of the video")
-                TRANSLATE_AUDIO_TO = gr.Dropdown(['English (en)', 'French (fr)', 'German (de)', 'Spanish (es)', 'Italian (it)', 'Japanese (ja)', 'Chinese (zh)', 'Dutch (nl)', 'Ukrainian (uk)', 'Portuguese (pt)'], value='English (en)',label = 'Translate audio to', info="Select the target language, and make sure to select the language corresponding to the speakers of the target language to avoid errors in the process.")
+                #video_input = gr.UploadButton("Click to Upload a video", file_types=["video"], file_count="single") #gr.Video() # height=300,width=300
+                video_input = gr.File(label="VIDEO")
+                #link = gr.HTML()
+                #video_input.change(submit_file_func, video_input, [video_input, link], show_progress='full')
+
+                SOURCE_LANGUAGE = gr.Dropdown(['Automatic detection', 'Arabic (ar)', 'Chinese (zh)', 'Czech (cs)', 'Danish (da)', 'Dutch (nl)', 'English (en)', 'Finnish (fi)', 'French (fr)', 'German (de)', 'Greek (el)', 'Hebrew (he)', 'Hindi (hi)', 'Hungarian (hu)', 'Italian (it)', 'Japanese (ja)', 'Korean (ko)', 'Persian (fa)', 'Polish (pl)', 'Portuguese (pt)', 'Russian (ru)', 'Spanish (es)', 'Turkish (tr)', 'Ukrainian (uk)', 'Urdu (ur)', 'Vietnamese (vi)'], value='Automatic detection',label = 'Source language', info="This is the original language of the video")
+                TRANSLATE_AUDIO_TO = gr.Dropdown(['Arabic (ar)', 'Chinese (zh)', 'Czech (cs)', 'Danish (da)', 'Dutch (nl)', 'English (en)', 'Finnish (fi)', 'French (fr)', 'German (de)', 'Greek (el)', 'Hebrew (he)', 'Hindi (hi)', 'Hungarian (hu)', 'Italian (it)', 'Japanese (ja)', 'Korean (ko)', 'Persian (fa)', 'Polish (pl)', 'Portuguese (pt)', 'Russian (ru)', 'Spanish (es)', 'Turkish (tr)', 'Ukrainian (uk)', 'Urdu (ur)', 'Vietnamese (vi)'], value='English (en)',label = 'Translate audio to', info="Select the target language, and make sure to select the language corresponding to the speakers of the target language to avoid errors in the process.")
 
                 line_ = gr.HTML("<hr></h2>")
                 gr.Markdown("Select how many people are speaking in the video.")
@@ -389,15 +687,15 @@ with gr.Blocks(theme=theme) as demo:
 
                 with gr.Column():
                       with gr.Accordion("Advanced Settings", open=False):
-                        
+
                           AUDIO_MIX = gr.Dropdown(['Mixing audio with sidechain compression', 'Adjusting volumes and mixing audio'], value='Adjusting volumes and mixing audio', label = 'Audio Mixing Method', info="Mix original and translated audio files to create a customized, balanced output with two available mixing modes.")
-                          
+
                           gr.HTML("<hr></h2>")
                           gr.Markdown("Default configuration of Whisper.")
                           WHISPER_MODEL_SIZE = gr.inputs.Dropdown(['tiny', 'base', 'small', 'medium', 'large-v1', 'large-v2'], default=whisper_model_default, label="Whisper model")
                           batch_size = gr.inputs.Slider(1, 32, default=16, label="Batch size", step=1)
                           compute_type = gr.inputs.Dropdown(list_compute_type, default=compute_type_default, label="Compute type")
-                          
+
                           gr.HTML("<hr></h2>")
                           VIDEO_OUTPUT_NAME = gr.Textbox(label="Translated file name" ,value="video_output.mp4", info="The name of the output file")
                           PREVIEW = gr.Checkbox(label="Preview", info="Preview cuts the video to only 10 seconds for testing purposes. Please deactivate it to retrieve the full video duration.")
@@ -406,7 +704,7 @@ with gr.Blocks(theme=theme) as demo:
                 with gr.Row():
                     video_button = gr.Button("TRANSLATE", )
                 with gr.Row():
-                    video_output = gr.Video()
+                    video_output = gr.outputs.File(label="DOWNLOAD TRANSLATED VIDEO") #gr.Video()
 
                 line_ = gr.HTML("<hr></h2>")
                 if os.getenv("YOUR_HF_TOKEN") == None or os.getenv("YOUR_HF_TOKEN") == "":
@@ -419,10 +717,10 @@ with gr.Blocks(theme=theme) as demo:
                         [
                             "./assets/Video_main.mp4",
                             "",
-                            True,
-                            "base",
+                            False,
+                            "large-v2",
                             16,
-                            "float32",
+                            "float16",
                             "Spanish (es)",
                             "English (en)",
                             1,
@@ -459,49 +757,19 @@ with gr.Blocks(theme=theme) as demo:
                     AUDIO_MIX,
                     ],
                     outputs=[video_output],
-                    cache_examples=True,
+                    cache_examples=False,
                 )
 
 ### link
 
-    with gr.Tab("Translate audio from video link"):
+    with gr.Tab("Audio Translation via Video Link"):
         with gr.Row():
             with gr.Column():
 
                 blink_input = gr.Textbox(label="Media link.", info="Example: www.youtube.com/watch?v=g_9rPvbENUw", placeholder="URL goes here...")
-                # bSOURCE_LANGUAGE = gr.Dropdown(['Automatic detection', 'en', 'fr', 'de', 'es', 'it', 'ja', 'zh', 'nl', 'uk', 'pt'], value='en',label = 'Source language')
-
-                # gr.HTML("<hr></h2>")
-
-                # bHFKEY = gr.Textbox(label="HF Token", info="One important step is to accept the license agreement for using Pyannote. You need to have an account on Hugging Face and accept the license to use the models: https://huggingface.co/pyannote/speaker-diarization and https://huggingface.co/pyannote/segmentation. Get your KEY TOKEN here: https://hf.co/settings/tokens", placeholder="Token goes here...")
-
-                # gr.Markdown("Select the target language, and make sure to select the language corresponding to the speakers of the target language to avoid errors in the process.")
-                # bTRANSLATE_AUDIO_TO = gr.inputs.Dropdown(['en', 'fr', 'de', 'es', 'it', 'ja', 'zh', 'nl', 'uk', 'pt'], default='en',label = 'Translate audio to')
-
-                # gr.Markdown("Select how many people are speaking in the video.")
-                # bmin_speakers = gr.inputs.Slider(1, 6, default=1, label="min_speakers", step=1, )
-                # bmax_speakers = gr.inputs.Slider(1, 6, default=2, label="max_speakers",step=1)
 
-                # gr.Markdown("Select the voice you want for each speaker.")
-                # btts_voice00 = gr.inputs.Dropdown(list_tts, default='en-AU-WilliamNeural-Male', label = 'TTS Speaker 1')
-                # btts_voice01 = gr.inputs.Dropdown(list_tts, default='en-CA-ClaraNeural-Female', label = 'TTS Speaker 2')
-                # btts_voice02 = gr.inputs.Dropdown(list_tts, default='en-GB-ThomasNeural-Male', label = 'TTS Speaker 3')
-                # btts_voice03 = gr.inputs.Dropdown(list_tts, default='en-GB-SoniaNeural-Female', label = 'TTS Speaker 4')
-                # btts_voice04 = gr.inputs.Dropdown(list_tts, default='en-NZ-MitchellNeural-Male', label = 'TTS Speaker 5')
-                # btts_voice05 = gr.inputs.Dropdown(list_tts, default='en-GB-MaisieNeural-Female', label = 'TTS Speaker 6')
-
-                # with gr.Column():
-                #       with gr.Accordion("Advanced Settings", open=False):
-                #           gr.Markdown("Default configuration of Whisper.")
-                #           bWHISPER_MODEL_SIZE = gr.inputs.Dropdown(['tiny', 'base', 'small', 'medium', 'large-v1', 'large-v2'], default=whisper_model_default, label="Whisper model")
-                #           bbatch_size = gr.inputs.Slider(1, 32, default=16, label="Batch size", step=1)
-                #           bcompute_type = gr.inputs.Dropdown(list_compute_type, default=compute_type_default, label="Compute type")
-
-                #           bPREVIEW = gr.inputs.Checkbox(label="Preview cuts the video to only 10 seconds for testing purposes. Please deactivate it to retrieve the full video duration.")
-                #           bVIDEO_OUTPUT_NAME = gr.Textbox(label="Translated file name" ,value="video_output.mp4")
-
-                bSOURCE_LANGUAGE = gr.Dropdown(['Automatic detection', 'English (en)', 'French (fr)', 'German (de)', 'Spanish (es)', 'Italian (it)', 'Japanese (ja)', 'Chinese (zh)', 'Dutch (nl)', 'Ukrainian (uk)', 'Portuguese (pt)'], value='Automatic detection',label = 'Source language', info="This is the original language of the video")
-                bTRANSLATE_AUDIO_TO = gr.Dropdown(['English (en)', 'French (fr)', 'German (de)', 'Spanish (es)', 'Italian (it)', 'Japanese (ja)', 'Chinese (zh)', 'Dutch (nl)', 'Ukrainian (uk)', 'Portuguese (pt)'], value='English (en)',label = 'Translate audio to', info="Select the target language, and make sure to select the language corresponding to the speakers of the target language to avoid errors in the process.")
+                bSOURCE_LANGUAGE = gr.Dropdown(['Automatic detection', 'Arabic (ar)', 'Chinese (zh)', 'Czech (cs)', 'Danish (da)', 'Dutch (nl)', 'English (en)', 'Finnish (fi)', 'French (fr)', 'German (de)', 'Greek (el)', 'Hebrew (he)', 'Hindi (hi)', 'Hungarian (hu)', 'Italian (it)', 'Japanese (ja)', 'Korean (ko)', 'Persian (fa)', 'Polish (pl)', 'Portuguese (pt)', 'Russian (ru)', 'Spanish (es)', 'Turkish (tr)', 'Ukrainian (uk)', 'Urdu (ur)', 'Vietnamese (vi)'], value='Automatic detection',label = 'Source language', info="This is the original language of the video")
+                bTRANSLATE_AUDIO_TO = gr.Dropdown(['Arabic (ar)', 'Chinese (zh)', 'Czech (cs)', 'Danish (da)', 'Dutch (nl)', 'English (en)', 'Finnish (fi)', 'French (fr)', 'German (de)', 'Greek (el)', 'Hebrew (he)', 'Hindi (hi)', 'Hungarian (hu)', 'Italian (it)', 'Japanese (ja)', 'Korean (ko)', 'Persian (fa)', 'Polish (pl)', 'Portuguese (pt)', 'Russian (ru)', 'Spanish (es)', 'Turkish (tr)', 'Ukrainian (uk)', 'Urdu (ur)', 'Vietnamese (vi)'], value='English (en)',label = 'Translate audio to', info="Select the target language, and make sure to select the language corresponding to the speakers of the target language to avoid errors in the process.")
 
                 bline_ = gr.HTML("<hr></h2>")
                 gr.Markdown("Select how many people are speaking in the video.")
@@ -524,7 +792,7 @@ with gr.Blocks(theme=theme) as demo:
 
                 with gr.Column():
                       with gr.Accordion("Advanced Settings", open=False):
-                          
+
                           bAUDIO_MIX = gr.Dropdown(['Mixing audio with sidechain compression', 'Adjusting volumes and mixing audio'], value='Adjusting volumes and mixing audio', label = 'Audio Mixing Method', info="Mix original and translated audio files to create a customized, balanced output with two available mixing modes.")
 
                           gr.HTML("<hr></h2>")
@@ -537,18 +805,11 @@ with gr.Blocks(theme=theme) as demo:
                           bVIDEO_OUTPUT_NAME = gr.Textbox(label="Translated file name" ,value="video_output.mp4", info="The name of the output file")
                           bPREVIEW = gr.Checkbox(label="Preview", info="Preview cuts the video to only 10 seconds for testing purposes. Please deactivate it to retrieve the full video duration.")
 
-
-
-                # text_button = gr.Button("Translate audio of video")
-                # link_output = gr.Video() #gr.outputs.File(label="Download!")
-
-
-
             with gr.Column(variant='compact'):
                 with gr.Row():
                     text_button = gr.Button("TRANSLATE")
                 with gr.Row():
-                    blink_output = gr.Video() #gr.outputs.File(label="Download!") # gr.Video()
+                    blink_output = gr.outputs.File(label="DOWNLOAD TRANSLATED VIDEO") # gr.Video()
 
 
                 bline_ = gr.HTML("<hr></h2>")
@@ -562,10 +823,10 @@ with gr.Blocks(theme=theme) as demo:
                         [
                             "https://www.youtube.com/watch?v=5ZeHtRKHl7Y",
                             "",
-                            True,
-                            "base",
+                            False,
+                            "large-v2",
                             16,
-                            "float32",
+                            "float16",
                             "Japanese (ja)",
                             "English (en)",
                             1,
@@ -602,15 +863,139 @@ with gr.Blocks(theme=theme) as demo:
                     bAUDIO_MIX
                     ],
                     outputs=[blink_output],
-                    cache_examples=True,
+                    cache_examples=False,
                 )
 
 
+    with gr.Tab("Custom voice RVC"):
+        with gr.Column():
+          with gr.Accordion("Download RVC Models", open=True):
+            url_links = gr.Textbox(label="URLs", value="",info="Automatically download the RVC models from the URL. You can use links from HuggingFace or Drive, and you can include several links, each one separated by a comma.", placeholder="urls here...", lines=1)
+            download_finish = gr.HTML()
+            download_button = gr.Button("DOWNLOAD MODELS")
+
+            def update_models():
+              models, index_paths = upload_model_list()
+              for i in range(8):                      
+                dict_models = {
+                    f'model_voice_path{i:02d}': gr.update(choices=models) for i in range(8)
+                }
+                dict_index = {
+                    f'file_index2_{i:02d}': gr.update(choices=index_paths) for i in range(8)
+                }
+                dict_changes = {**dict_models, **dict_index}
+                return [value for value in dict_changes.values()]
+
+        with gr.Column():
+          with gr.Accordion("Replace voice: TTS to RVC", open=False):
+            with gr.Column(variant='compact'):
+              with gr.Column():
+                gr.Markdown("### 1. To enable its use, mark it as enable.")
+                enable_custom_voice = gr.Checkbox(label="ENABLE", info="Check this to enable the use of the models.")
+                enable_custom_voice.change(custom_model_voice_enable, [enable_custom_voice], [])
+
+                gr.Markdown("### 2. Select a voice that will be applied to each TTS of each corresponding speaker and apply the configurations.")
+                
+                gr.Markdown("Voice to apply to the first speaker.")
+                with gr.Row():
+                  model_voice_path00 = gr.Dropdown(models, label = 'Model-1', visible=True, interactive= True)
+                  file_index2_00 = gr.Dropdown(index_paths, label = 'Index-1', visible=True, interactive= True)
+                  name_transpose00 = gr.Number(label = 'Transpose-1', value=0, visible=True, interactive= True)
+                gr.HTML("<hr></h2>")
+                gr.Markdown("Voice to apply to the second speaker.")
+                with gr.Row():
+                  model_voice_path01 = gr.Dropdown(models, label='Model-2', visible=True, interactive=True)
+                  file_index2_01 = gr.Dropdown(index_paths, label='Index-2', visible=True, interactive=True)
+                  name_transpose01 = gr.Number(label='Transpose-2', value=0, visible=True, interactive=True)
+                gr.HTML("<hr></h2>")
+                gr.Markdown("Voice to apply to the third speaker.")
+                with gr.Row():
+                  model_voice_path02 = gr.Dropdown(models, label='Model-3', visible=True, interactive=True)
+                  file_index2_02 = gr.Dropdown(index_paths, label='Index-3', visible=True, interactive=True)
+                  name_transpose02 = gr.Number(label='Transpose-3', value=0, visible=True, interactive=True)
+                gr.HTML("<hr></h2>")
+                gr.Markdown("Voice to apply to the fourth speaker.")
+                with gr.Row():
+                  model_voice_path03 = gr.Dropdown(models, label='Model-4', visible=True, interactive=True)
+                  file_index2_03 = gr.Dropdown(index_paths, label='Index-4', visible=True, interactive=True)
+                  name_transpose03 = gr.Number(label='Transpose-4', value=0, visible=True, interactive=True)
+                gr.HTML("<hr></h2>")
+                gr.Markdown("Voice to apply to the fifth speaker.")
+                with gr.Row():
+                  model_voice_path04 = gr.Dropdown(models, label='Model-5', visible=True, interactive=True)
+                  file_index2_04 = gr.Dropdown(index_paths, label='Index-5', visible=True, interactive=True)
+                  name_transpose04 = gr.Number(label='Transpose-5', value=0, visible=True, interactive=True)
+                gr.HTML("<hr></h2>")
+                gr.Markdown("Voice to apply to the sixth speaker.")
+                with gr.Row():
+                  model_voice_path05 = gr.Dropdown(models, label='Model-6', visible=True, interactive=True)
+                  file_index2_05 = gr.Dropdown(index_paths, label='Index-6', visible=True, interactive=True)
+                  name_transpose05 = gr.Number(label='Transpose-6', value=0, visible=True, interactive=True)
+                gr.HTML("<hr></h2>")
+                gr.Markdown("- Voice to apply in case a speaker is not detected successfully.")
+                with gr.Row():
+                  model_voice_path06 = gr.Dropdown(models, label='Model-Aux', visible=True, interactive=True)
+                  file_index2_06 = gr.Dropdown(index_paths, label='Index-Aux', visible=True, interactive=True)
+                  name_transpose06 = gr.Number(label='Transpose-Aux', value=0, visible=True, interactive=True)
+                gr.HTML("<hr></h2>")
+                with gr.Row():
+                  f0_method_global = gr.Dropdown(f0_methods_voice, value='pm', label = 'Global F0 method', visible=True, interactive= True)
+
+            with gr.Row(variant='compact'):
+              button_config = gr.Button("APPLY CONFIGURATION")
+
+              confirm_conf = gr.HTML()
+
+            button_config.click(voices.apply_conf, inputs=[
+                f0_method_global,
+                model_voice_path00, name_transpose00, file_index2_00,
+                model_voice_path01, name_transpose01, file_index2_01,
+                model_voice_path02, name_transpose02, file_index2_02,
+                model_voice_path03, name_transpose03, file_index2_03,
+                model_voice_path04, name_transpose04, file_index2_04,
+                model_voice_path05, name_transpose05, file_index2_05,
+                model_voice_path06, name_transpose06, file_index2_06,
+                ], outputs=[confirm_conf])
+
+
+          with gr.Column():
+                with gr.Accordion("Test RVC", open=False):
+
+                  with gr.Row(variant='compact'):
+                    text_test = gr.Textbox(label="Text", value="This is an example",info="write a text", placeholder="...", lines=5)
+                    with gr.Column(): 
+                      tts_test = gr.Dropdown(list_tts, value='en-GB-ThomasNeural-Male', label = 'TTS', visible=True, interactive= True)
+                      model_voice_path07 = gr.Dropdown(models, label = 'Model', visible=True, interactive= True) #value=''
+                      file_index2_07 = gr.Dropdown(index_paths, label = 'Index', visible=True, interactive= True) #value=''
+                      transpose_test = gr.Number(label = 'Transpose', value=0, visible=True, interactive= True, info="integer, number of semitones, raise by an octave: 12, lower by an octave: -12")
+                      f0method_test = gr.Dropdown(f0_methods_voice, value='pm', label = 'F0 method', visible=True, interactive= True) 
+                  with gr.Row(variant='compact'):
+                    button_test = gr.Button("Test audio")
+
+                  with gr.Column():
+                    with gr.Row():
+                      original_ttsvoice = gr.Audio()
+                      ttsvoice = gr.Audio()
+
+                    button_test.click(voices.make_test, inputs=[
+                        text_test,
+                        tts_test,
+                        model_voice_path07,
+                        file_index2_07,
+                        transpose_test,
+                        f0method_test,
+                        ], outputs=[ttsvoice, original_ttsvoice])
+
+                download_button.click(download_list, [url_links], [download_finish]).then(update_models, [], 
+                                  [
+                                    model_voice_path00, model_voice_path01, model_voice_path02, model_voice_path03, model_voice_path04, model_voice_path05, model_voice_path06, model_voice_path07,
+                                    file_index2_00, file_index2_01, file_index2_02, file_index2_03, file_index2_04, file_index2_05, file_index2_06, file_index2_07
+                                  ])
 
 
     with gr.Tab("Help"):
-        gr.Markdown(news)
         gr.Markdown(tutorial)
+        gr.Markdown(news)
 
     with gr.Accordion("Logs", open = False):
         logs = gr.Textbox()
@@ -658,5 +1043,5 @@ with gr.Blocks(theme=theme) as demo:
         bAUDIO_MIX,
         ], outputs=blink_output)
 
-demo.launch(enable_queue=True)
-#demo.launch()
+#demo.launch(debug=True, enable_queue=True)
+demo.launch(share=True, enable_queue=True, quiet=True, debug=False)

configs/32k.json

@@ -0,0 +1,46 @@
+{
+  "train": {
+    "log_interval": 200,
+    "seed": 1234,
+    "epochs": 20000,
+    "learning_rate": 1e-4,
+    "betas": [0.8, 0.99],
+    "eps": 1e-9,
+    "batch_size": 4,
+    "fp16_run": false,
+    "lr_decay": 0.999875,
+    "segment_size": 12800,
+    "init_lr_ratio": 1,
+    "warmup_epochs": 0,
+    "c_mel": 45,
+    "c_kl": 1.0
+  },
+  "data": {
+    "max_wav_value": 32768.0,
+    "sampling_rate": 32000,
+    "filter_length": 1024,
+    "hop_length": 320,
+    "win_length": 1024,
+    "n_mel_channels": 80,
+    "mel_fmin": 0.0,
+    "mel_fmax": null
+  },
+  "model": {
+    "inter_channels": 192,
+    "hidden_channels": 192,
+    "filter_channels": 768,
+    "n_heads": 2,
+    "n_layers": 6,
+    "kernel_size": 3,
+    "p_dropout": 0,
+    "resblock": "1",
+    "resblock_kernel_sizes": [3,7,11],
+    "resblock_dilation_sizes": [[1,3,5], [1,3,5], [1,3,5]],
+    "upsample_rates": [10,4,2,2,2],
+    "upsample_initial_channel": 512,
+    "upsample_kernel_sizes": [16,16,4,4,4],
+    "use_spectral_norm": false,
+    "gin_channels": 256,
+    "spk_embed_dim": 109
+  }
+}

configs/32k_v2.json

@@ -0,0 +1,46 @@
+{
+  "train": {
+    "log_interval": 200,
+    "seed": 1234,
+    "epochs": 20000,
+    "learning_rate": 1e-4,
+    "betas": [0.8, 0.99],
+    "eps": 1e-9,
+    "batch_size": 4,
+    "fp16_run": false,
+    "lr_decay": 0.999875,
+    "segment_size": 12800,
+    "init_lr_ratio": 1,
+    "warmup_epochs": 0,
+    "c_mel": 45,
+    "c_kl": 1.0
+  },
+  "data": {
+    "max_wav_value": 32768.0,
+    "sampling_rate": 32000,
+    "filter_length": 1024,
+    "hop_length": 320,
+    "win_length": 1024,
+    "n_mel_channels": 80,
+    "mel_fmin": 0.0,
+    "mel_fmax": null
+  },
+  "model": {
+    "inter_channels": 192,
+    "hidden_channels": 192,
+    "filter_channels": 768,
+    "n_heads": 2,
+    "n_layers": 6,
+    "kernel_size": 3,
+    "p_dropout": 0,
+    "resblock": "1",
+    "resblock_kernel_sizes": [3,7,11],
+    "resblock_dilation_sizes": [[1,3,5], [1,3,5], [1,3,5]],
+    "upsample_rates": [10,8,2,2],
+    "upsample_initial_channel": 512,
+    "upsample_kernel_sizes": [20,16,4,4],
+    "use_spectral_norm": false,
+    "gin_channels": 256,
+    "spk_embed_dim": 109
+  }
+}

configs/40k.json

@@ -0,0 +1,46 @@
+{
+  "train": {
+    "log_interval": 200,
+    "seed": 1234,
+    "epochs": 20000,
+    "learning_rate": 1e-4,
+    "betas": [0.8, 0.99],
+    "eps": 1e-9,
+    "batch_size": 4,
+    "fp16_run": false,
+    "lr_decay": 0.999875,
+    "segment_size": 12800,
+    "init_lr_ratio": 1,
+    "warmup_epochs": 0,
+    "c_mel": 45,
+    "c_kl": 1.0
+  },
+  "data": {
+    "max_wav_value": 32768.0,
+    "sampling_rate": 40000,
+    "filter_length": 2048,
+    "hop_length": 400,
+    "win_length": 2048,
+    "n_mel_channels": 125,
+    "mel_fmin": 0.0,
+    "mel_fmax": null
+  },
+  "model": {
+    "inter_channels": 192,
+    "hidden_channels": 192,
+    "filter_channels": 768,
+    "n_heads": 2,
+    "n_layers": 6,
+    "kernel_size": 3,
+    "p_dropout": 0,
+    "resblock": "1",
+    "resblock_kernel_sizes": [3,7,11],
+    "resblock_dilation_sizes": [[1,3,5], [1,3,5], [1,3,5]],
+    "upsample_rates": [10,10,2,2],
+    "upsample_initial_channel": 512,
+    "upsample_kernel_sizes": [16,16,4,4],
+    "use_spectral_norm": false,
+    "gin_channels": 256,
+    "spk_embed_dim": 109
+  }
+}

configs/48k.json

@@ -0,0 +1,46 @@
+{
+  "train": {
+    "log_interval": 200,
+    "seed": 1234,
+    "epochs": 20000,
+    "learning_rate": 1e-4,
+    "betas": [0.8, 0.99],
+    "eps": 1e-9,
+    "batch_size": 4,
+    "fp16_run": false,
+    "lr_decay": 0.999875,
+    "segment_size": 11520,
+    "init_lr_ratio": 1,
+    "warmup_epochs": 0,
+    "c_mel": 45,
+    "c_kl": 1.0
+  },
+  "data": {
+    "max_wav_value": 32768.0,
+    "sampling_rate": 48000,
+    "filter_length": 2048,
+    "hop_length": 480,
+    "win_length": 2048,
+    "n_mel_channels": 128,
+    "mel_fmin": 0.0,
+    "mel_fmax": null
+  },
+  "model": {
+    "inter_channels": 192,
+    "hidden_channels": 192,
+    "filter_channels": 768,
+    "n_heads": 2,
+    "n_layers": 6,
+    "kernel_size": 3,
+    "p_dropout": 0,
+    "resblock": "1",
+    "resblock_kernel_sizes": [3,7,11],
+    "resblock_dilation_sizes": [[1,3,5], [1,3,5], [1,3,5]],
+    "upsample_rates": [10,6,2,2,2],
+    "upsample_initial_channel": 512,
+    "upsample_kernel_sizes": [16,16,4,4,4],
+    "use_spectral_norm": false,
+    "gin_channels": 256,
+    "spk_embed_dim": 109
+  }
+}

configs/48k_v2.json

@@ -0,0 +1,46 @@
+{
+  "train": {
+    "log_interval": 200,
+    "seed": 1234,
+    "epochs": 20000,
+    "learning_rate": 1e-4,
+    "betas": [0.8, 0.99],
+    "eps": 1e-9,
+    "batch_size": 4,
+    "fp16_run": false,
+    "lr_decay": 0.999875,
+    "segment_size": 17280,
+    "init_lr_ratio": 1,
+    "warmup_epochs": 0,
+    "c_mel": 45,
+    "c_kl": 1.0
+  },
+  "data": {
+    "max_wav_value": 32768.0,
+    "sampling_rate": 48000,
+    "filter_length": 2048,
+    "hop_length": 480,
+    "win_length": 2048,
+    "n_mel_channels": 128,
+    "mel_fmin": 0.0,
+    "mel_fmax": null
+  },
+  "model": {
+    "inter_channels": 192,
+    "hidden_channels": 192,
+    "filter_channels": 768,
+    "n_heads": 2,
+    "n_layers": 6,
+    "kernel_size": 3,
+    "p_dropout": 0,
+    "resblock": "1",
+    "resblock_kernel_sizes": [3,7,11],
+    "resblock_dilation_sizes": [[1,3,5], [1,3,5], [1,3,5]],
+    "upsample_rates": [12,10,2,2],
+    "upsample_initial_channel": 512,
+    "upsample_kernel_sizes": [24,20,4,4],
+    "use_spectral_norm": false,
+    "gin_channels": 256,
+    "spk_embed_dim": 109
+  }
+}

lib/audio.py

@@ -0,0 +1,21 @@
+import ffmpeg
+import numpy as np
+
+
+def load_audio(file, sr):
+    try:
+        # https://github.com/openai/whisper/blob/main/whisper/audio.py#L26
+        # This launches a subprocess to decode audio while down-mixing and resampling as necessary.
+        # Requires the ffmpeg CLI and `ffmpeg-python` package to be installed.
+        file = (
+            file.strip(" ").strip('"').strip("\n").strip('"').strip(" ")
+        )  # To prevent beginners from copying paths with leading or trailing spaces, quotation marks, and line breaks.
+        out, _ = (
+            ffmpeg.input(file, threads=0)
+            .output("-", format="f32le", acodec="pcm_f32le", ac=1, ar=sr)
+            .run(cmd=["ffmpeg", "-nostdin"], capture_stdout=True, capture_stderr=True)
+        )
+    except Exception as e:
+        raise RuntimeError(f"Failed to load audio: {e}")
+
+    return np.frombuffer(out, np.float32).flatten()

lib/infer_pack/attentions.py

@@ -0,0 +1,417 @@
+import copy
+import math
+import numpy as np
+import torch
+from torch import nn
+from torch.nn import functional as F
+
+from lib.infer_pack import commons
+from lib.infer_pack import modules
+from lib.infer_pack.modules import LayerNorm
+
+
+class Encoder(nn.Module):
+    def __init__(
+        self,
+        hidden_channels,
+        filter_channels,
+        n_heads,
+        n_layers,
+        kernel_size=1,
+        p_dropout=0.0,
+        window_size=10,
+        **kwargs
+    ):
+        super().__init__()
+        self.hidden_channels = hidden_channels
+        self.filter_channels = filter_channels
+        self.n_heads = n_heads
+        self.n_layers = n_layers
+        self.kernel_size = kernel_size
+        self.p_dropout = p_dropout
+        self.window_size = window_size
+
+        self.drop = nn.Dropout(p_dropout)
+        self.attn_layers = nn.ModuleList()
+        self.norm_layers_1 = nn.ModuleList()
+        self.ffn_layers = nn.ModuleList()
+        self.norm_layers_2 = nn.ModuleList()
+        for i in range(self.n_layers):
+            self.attn_layers.append(
+                MultiHeadAttention(
+                    hidden_channels,
+                    hidden_channels,
+                    n_heads,
+                    p_dropout=p_dropout,
+                    window_size=window_size,
+                )
+            )
+            self.norm_layers_1.append(LayerNorm(hidden_channels))
+            self.ffn_layers.append(
+                FFN(
+                    hidden_channels,
+                    hidden_channels,
+                    filter_channels,
+                    kernel_size,
+                    p_dropout=p_dropout,
+                )
+            )
+            self.norm_layers_2.append(LayerNorm(hidden_channels))
+
+    def forward(self, x, x_mask):
+        attn_mask = x_mask.unsqueeze(2) * x_mask.unsqueeze(-1)
+        x = x * x_mask
+        for i in range(self.n_layers):
+            y = self.attn_layers[i](x, x, attn_mask)
+            y = self.drop(y)
+            x = self.norm_layers_1[i](x + y)
+
+            y = self.ffn_layers[i](x, x_mask)
+            y = self.drop(y)
+            x = self.norm_layers_2[i](x + y)
+        x = x * x_mask
+        return x
+
+
+class Decoder(nn.Module):
+    def __init__(
+        self,
+        hidden_channels,
+        filter_channels,
+        n_heads,
+        n_layers,
+        kernel_size=1,
+        p_dropout=0.0,
+        proximal_bias=False,
+        proximal_init=True,
+        **kwargs
+    ):
+        super().__init__()
+        self.hidden_channels = hidden_channels
+        self.filter_channels = filter_channels
+        self.n_heads = n_heads
+        self.n_layers = n_layers
+        self.kernel_size = kernel_size
+        self.p_dropout = p_dropout
+        self.proximal_bias = proximal_bias
+        self.proximal_init = proximal_init
+
+        self.drop = nn.Dropout(p_dropout)
+        self.self_attn_layers = nn.ModuleList()
+        self.norm_layers_0 = nn.ModuleList()
+        self.encdec_attn_layers = nn.ModuleList()
+        self.norm_layers_1 = nn.ModuleList()
+        self.ffn_layers = nn.ModuleList()
+        self.norm_layers_2 = nn.ModuleList()
+        for i in range(self.n_layers):
+            self.self_attn_layers.append(
+                MultiHeadAttention(
+                    hidden_channels,
+                    hidden_channels,
+                    n_heads,
+                    p_dropout=p_dropout,
+                    proximal_bias=proximal_bias,
+                    proximal_init=proximal_init,
+                )
+            )
+            self.norm_layers_0.append(LayerNorm(hidden_channels))
+            self.encdec_attn_layers.append(
+                MultiHeadAttention(
+                    hidden_channels, hidden_channels, n_heads, p_dropout=p_dropout
+                )
+            )
+            self.norm_layers_1.append(LayerNorm(hidden_channels))
+            self.ffn_layers.append(
+                FFN(
+                    hidden_channels,
+                    hidden_channels,
+                    filter_channels,
+                    kernel_size,
+                    p_dropout=p_dropout,
+                    causal=True,
+                )
+            )
+            self.norm_layers_2.append(LayerNorm(hidden_channels))
+
+    def forward(self, x, x_mask, h, h_mask):
+        """
+        x: decoder input
+        h: encoder output
+        """
+        self_attn_mask = commons.subsequent_mask(x_mask.size(2)).to(
+            device=x.device, dtype=x.dtype
+        )
+        encdec_attn_mask = h_mask.unsqueeze(2) * x_mask.unsqueeze(-1)
+        x = x * x_mask
+        for i in range(self.n_layers):
+            y = self.self_attn_layers[i](x, x, self_attn_mask)
+            y = self.drop(y)
+            x = self.norm_layers_0[i](x + y)
+
+            y = self.encdec_attn_layers[i](x, h, encdec_attn_mask)
+            y = self.drop(y)
+            x = self.norm_layers_1[i](x + y)
+
+            y = self.ffn_layers[i](x, x_mask)
+            y = self.drop(y)
+            x = self.norm_layers_2[i](x + y)
+        x = x * x_mask
+        return x
+
+
+class MultiHeadAttention(nn.Module):
+    def __init__(
+        self,
+        channels,
+        out_channels,
+        n_heads,
+        p_dropout=0.0,
+        window_size=None,
+        heads_share=True,
+        block_length=None,
+        proximal_bias=False,
+        proximal_init=False,
+    ):
+        super().__init__()
+        assert channels % n_heads == 0
+
+        self.channels = channels
+        self.out_channels = out_channels
+        self.n_heads = n_heads
+        self.p_dropout = p_dropout
+        self.window_size = window_size
+        self.heads_share = heads_share
+        self.block_length = block_length
+        self.proximal_bias = proximal_bias
+        self.proximal_init = proximal_init
+        self.attn = None
+
+        self.k_channels = channels // n_heads
+        self.conv_q = nn.Conv1d(channels, channels, 1)
+        self.conv_k = nn.Conv1d(channels, channels, 1)
+        self.conv_v = nn.Conv1d(channels, channels, 1)
+        self.conv_o = nn.Conv1d(channels, out_channels, 1)
+        self.drop = nn.Dropout(p_dropout)
+
+        if window_size is not None:
+            n_heads_rel = 1 if heads_share else n_heads
+            rel_stddev = self.k_channels**-0.5
+            self.emb_rel_k = nn.Parameter(
+                torch.randn(n_heads_rel, window_size * 2 + 1, self.k_channels)
+                * rel_stddev
+            )
+            self.emb_rel_v = nn.Parameter(
+                torch.randn(n_heads_rel, window_size * 2 + 1, self.k_channels)
+                * rel_stddev
+            )
+
+        nn.init.xavier_uniform_(self.conv_q.weight)
+        nn.init.xavier_uniform_(self.conv_k.weight)
+        nn.init.xavier_uniform_(self.conv_v.weight)
+        if proximal_init:
+            with torch.no_grad():
+                self.conv_k.weight.copy_(self.conv_q.weight)
+                self.conv_k.bias.copy_(self.conv_q.bias)
+
+    def forward(self, x, c, attn_mask=None):
+        q = self.conv_q(x)
+        k = self.conv_k(c)
+        v = self.conv_v(c)
+
+        x, self.attn = self.attention(q, k, v, mask=attn_mask)
+
+        x = self.conv_o(x)
+        return x
+
+    def attention(self, query, key, value, mask=None):
+        # reshape [b, d, t] -> [b, n_h, t, d_k]
+        b, d, t_s, t_t = (*key.size(), query.size(2))
+        query = query.view(b, self.n_heads, self.k_channels, t_t).transpose(2, 3)
+        key = key.view(b, self.n_heads, self.k_channels, t_s).transpose(2, 3)
+        value = value.view(b, self.n_heads, self.k_channels, t_s).transpose(2, 3)
+
+        scores = torch.matmul(query / math.sqrt(self.k_channels), key.transpose(-2, -1))
+        if self.window_size is not None:
+            assert (
+                t_s == t_t
+            ), "Relative attention is only available for self-attention."
+            key_relative_embeddings = self._get_relative_embeddings(self.emb_rel_k, t_s)
+            rel_logits = self._matmul_with_relative_keys(
+                query / math.sqrt(self.k_channels), key_relative_embeddings
+            )
+            scores_local = self._relative_position_to_absolute_position(rel_logits)
+            scores = scores + scores_local
+        if self.proximal_bias:
+            assert t_s == t_t, "Proximal bias is only available for self-attention."
+            scores = scores + self._attention_bias_proximal(t_s).to(
+                device=scores.device, dtype=scores.dtype
+            )
+        if mask is not None:
+            scores = scores.masked_fill(mask == 0, -1e4)
+            if self.block_length is not None:
+                assert (
+                    t_s == t_t
+                ), "Local attention is only available for self-attention."
+                block_mask = (
+                    torch.ones_like(scores)
+                    .triu(-self.block_length)
+                    .tril(self.block_length)
+                )
+                scores = scores.masked_fill(block_mask == 0, -1e4)
+        p_attn = F.softmax(scores, dim=-1)  # [b, n_h, t_t, t_s]
+        p_attn = self.drop(p_attn)
+        output = torch.matmul(p_attn, value)
+        if self.window_size is not None:
+            relative_weights = self._absolute_position_to_relative_position(p_attn)
+            value_relative_embeddings = self._get_relative_embeddings(
+                self.emb_rel_v, t_s
+            )
+            output = output + self._matmul_with_relative_values(
+                relative_weights, value_relative_embeddings
+            )
+        output = (
+            output.transpose(2, 3).contiguous().view(b, d, t_t)
+        )  # [b, n_h, t_t, d_k] -> [b, d, t_t]
+        return output, p_attn
+
+    def _matmul_with_relative_values(self, x, y):
+        """
+        x: [b, h, l, m]
+        y: [h or 1, m, d]
+        ret: [b, h, l, d]
+        """
+        ret = torch.matmul(x, y.unsqueeze(0))
+        return ret
+
+    def _matmul_with_relative_keys(self, x, y):
+        """
+        x: [b, h, l, d]
+        y: [h or 1, m, d]
+        ret: [b, h, l, m]
+        """
+        ret = torch.matmul(x, y.unsqueeze(0).transpose(-2, -1))
+        return ret
+
+    def _get_relative_embeddings(self, relative_embeddings, length):
+        max_relative_position = 2 * self.window_size + 1
+        # Pad first before slice to avoid using cond ops.
+        pad_length = max(length - (self.window_size + 1), 0)
+        slice_start_position = max((self.window_size + 1) - length, 0)
+        slice_end_position = slice_start_position + 2 * length - 1
+        if pad_length > 0:
+            padded_relative_embeddings = F.pad(
+                relative_embeddings,
+                commons.convert_pad_shape([[0, 0], [pad_length, pad_length], [0, 0]]),
+            )
+        else:
+            padded_relative_embeddings = relative_embeddings
+        used_relative_embeddings = padded_relative_embeddings[
+            :, slice_start_position:slice_end_position
+        ]
+        return used_relative_embeddings
+
+    def _relative_position_to_absolute_position(self, x):
+        """
+        x: [b, h, l, 2*l-1]
+        ret: [b, h, l, l]
+        """
+        batch, heads, length, _ = x.size()
+        # Concat columns of pad to shift from relative to absolute indexing.
+        x = F.pad(x, commons.convert_pad_shape([[0, 0], [0, 0], [0, 0], [0, 1]]))
+
+        # Concat extra elements so to add up to shape (len+1, 2*len-1).
+        x_flat = x.view([batch, heads, length * 2 * length])
+        x_flat = F.pad(
+            x_flat, commons.convert_pad_shape([[0, 0], [0, 0], [0, length - 1]])
+        )
+
+        # Reshape and slice out the padded elements.
+        x_final = x_flat.view([batch, heads, length + 1, 2 * length - 1])[
+            :, :, :length, length - 1 :
+        ]
+        return x_final
+
+    def _absolute_position_to_relative_position(self, x):
+        """
+        x: [b, h, l, l]
+        ret: [b, h, l, 2*l-1]
+        """
+        batch, heads, length, _ = x.size()
+        # padd along column
+        x = F.pad(
+            x, commons.convert_pad_shape([[0, 0], [0, 0], [0, 0], [0, length - 1]])
+        )
+        x_flat = x.view([batch, heads, length**2 + length * (length - 1)])
+        # add 0's in the beginning that will skew the elements after reshape
+        x_flat = F.pad(x_flat, commons.convert_pad_shape([[0, 0], [0, 0], [length, 0]]))
+        x_final = x_flat.view([batch, heads, length, 2 * length])[:, :, :, 1:]
+        return x_final
+
+    def _attention_bias_proximal(self, length):
+        """Bias for self-attention to encourage attention to close positions.
+        Args:
+          length: an integer scalar.
+        Returns:
+          a Tensor with shape [1, 1, length, length]
+        """
+        r = torch.arange(length, dtype=torch.float32)
+        diff = torch.unsqueeze(r, 0) - torch.unsqueeze(r, 1)
+        return torch.unsqueeze(torch.unsqueeze(-torch.log1p(torch.abs(diff)), 0), 0)
+
+
+class FFN(nn.Module):
+    def __init__(
+        self,
+        in_channels,
+        out_channels,
+        filter_channels,
+        kernel_size,
+        p_dropout=0.0,
+        activation=None,
+        causal=False,
+    ):
+        super().__init__()
+        self.in_channels = in_channels
+        self.out_channels = out_channels
+        self.filter_channels = filter_channels
+        self.kernel_size = kernel_size
+        self.p_dropout = p_dropout
+        self.activation = activation
+        self.causal = causal
+
+        if causal:
+            self.padding = self._causal_padding
+        else:
+            self.padding = self._same_padding
+
+        self.conv_1 = nn.Conv1d(in_channels, filter_channels, kernel_size)
+        self.conv_2 = nn.Conv1d(filter_channels, out_channels, kernel_size)
+        self.drop = nn.Dropout(p_dropout)
+
+    def forward(self, x, x_mask):
+        x = self.conv_1(self.padding(x * x_mask))
+        if self.activation == "gelu":
+            x = x * torch.sigmoid(1.702 * x)
+        else:
+            x = torch.relu(x)
+        x = self.drop(x)
+        x = self.conv_2(self.padding(x * x_mask))
+        return x * x_mask
+
+    def _causal_padding(self, x):
+        if self.kernel_size == 1:
+            return x
+        pad_l = self.kernel_size - 1
+        pad_r = 0
+        padding = [[0, 0], [0, 0], [pad_l, pad_r]]
+        x = F.pad(x, commons.convert_pad_shape(padding))
+        return x
+
+    def _same_padding(self, x):
+        if self.kernel_size == 1:
+            return x
+        pad_l = (self.kernel_size - 1) // 2
+        pad_r = self.kernel_size // 2
+        padding = [[0, 0], [0, 0], [pad_l, pad_r]]
+        x = F.pad(x, commons.convert_pad_shape(padding))
+        return x

lib/infer_pack/commons.py

@@ -0,0 +1,166 @@
+import math
+import numpy as np
+import torch
+from torch import nn
+from torch.nn import functional as F
+
+
+def init_weights(m, mean=0.0, std=0.01):
+    classname = m.__class__.__name__
+    if classname.find("Conv") != -1:
+        m.weight.data.normal_(mean, std)
+
+
+def get_padding(kernel_size, dilation=1):
+    return int((kernel_size * dilation - dilation) / 2)
+
+
+def convert_pad_shape(pad_shape):
+    l = pad_shape[::-1]
+    pad_shape = [item for sublist in l for item in sublist]
+    return pad_shape
+
+
+def kl_divergence(m_p, logs_p, m_q, logs_q):
+    """KL(P||Q)"""
+    kl = (logs_q - logs_p) - 0.5
+    kl += (
+        0.5 * (torch.exp(2.0 * logs_p) + ((m_p - m_q) ** 2)) * torch.exp(-2.0 * logs_q)
+    )
+    return kl
+
+
+def rand_gumbel(shape):
+    """Sample from the Gumbel distribution, protect from overflows."""
+    uniform_samples = torch.rand(shape) * 0.99998 + 0.00001
+    return -torch.log(-torch.log(uniform_samples))
+
+
+def rand_gumbel_like(x):
+    g = rand_gumbel(x.size()).to(dtype=x.dtype, device=x.device)
+    return g
+
+
+def slice_segments(x, ids_str, segment_size=4):
+    ret = torch.zeros_like(x[:, :, :segment_size])
+    for i in range(x.size(0)):
+        idx_str = ids_str[i]
+        idx_end = idx_str + segment_size
+        ret[i] = x[i, :, idx_str:idx_end]
+    return ret
+
+
+def slice_segments2(x, ids_str, segment_size=4):
+    ret = torch.zeros_like(x[:, :segment_size])
+    for i in range(x.size(0)):
+        idx_str = ids_str[i]
+        idx_end = idx_str + segment_size
+        ret[i] = x[i, idx_str:idx_end]
+    return ret
+
+
+def rand_slice_segments(x, x_lengths=None, segment_size=4):
+    b, d, t = x.size()
+    if x_lengths is None:
+        x_lengths = t
+    ids_str_max = x_lengths - segment_size + 1
+    ids_str = (torch.rand([b]).to(device=x.device) * ids_str_max).to(dtype=torch.long)
+    ret = slice_segments(x, ids_str, segment_size)
+    return ret, ids_str
+
+
+def get_timing_signal_1d(length, channels, min_timescale=1.0, max_timescale=1.0e4):
+    position = torch.arange(length, dtype=torch.float)
+    num_timescales = channels // 2
+    log_timescale_increment = math.log(float(max_timescale) / float(min_timescale)) / (
+        num_timescales - 1
+    )
+    inv_timescales = min_timescale * torch.exp(
+        torch.arange(num_timescales, dtype=torch.float) * -log_timescale_increment
+    )
+    scaled_time = position.unsqueeze(0) * inv_timescales.unsqueeze(1)
+    signal = torch.cat([torch.sin(scaled_time), torch.cos(scaled_time)], 0)
+    signal = F.pad(signal, [0, 0, 0, channels % 2])
+    signal = signal.view(1, channels, length)
+    return signal
+
+
+def add_timing_signal_1d(x, min_timescale=1.0, max_timescale=1.0e4):
+    b, channels, length = x.size()
+    signal = get_timing_signal_1d(length, channels, min_timescale, max_timescale)
+    return x + signal.to(dtype=x.dtype, device=x.device)
+
+
+def cat_timing_signal_1d(x, min_timescale=1.0, max_timescale=1.0e4, axis=1):
+    b, channels, length = x.size()
+    signal = get_timing_signal_1d(length, channels, min_timescale, max_timescale)
+    return torch.cat([x, signal.to(dtype=x.dtype, device=x.device)], axis)
+
+
+def subsequent_mask(length):
+    mask = torch.tril(torch.ones(length, length)).unsqueeze(0).unsqueeze(0)
+    return mask
+
+
+@torch.jit.script
+def fused_add_tanh_sigmoid_multiply(input_a, input_b, n_channels):
+    n_channels_int = n_channels[0]
+    in_act = input_a + input_b
+    t_act = torch.tanh(in_act[:, :n_channels_int, :])
+    s_act = torch.sigmoid(in_act[:, n_channels_int:, :])
+    acts = t_act * s_act
+    return acts
+
+
+def convert_pad_shape(pad_shape):
+    l = pad_shape[::-1]
+    pad_shape = [item for sublist in l for item in sublist]
+    return pad_shape
+
+
+def shift_1d(x):
+    x = F.pad(x, convert_pad_shape([[0, 0], [0, 0], [1, 0]]))[:, :, :-1]
+    return x
+
+
+def sequence_mask(length, max_length=None):
+    if max_length is None:
+        max_length = length.max()
+    x = torch.arange(max_length, dtype=length.dtype, device=length.device)
+    return x.unsqueeze(0) < length.unsqueeze(1)
+
+
+def generate_path(duration, mask):
+    """
+    duration: [b, 1, t_x]
+    mask: [b, 1, t_y, t_x]
+    """
+    device = duration.device
+
+    b, _, t_y, t_x = mask.shape
+    cum_duration = torch.cumsum(duration, -1)
+
+    cum_duration_flat = cum_duration.view(b * t_x)
+    path = sequence_mask(cum_duration_flat, t_y).to(mask.dtype)
+    path = path.view(b, t_x, t_y)
+    path = path - F.pad(path, convert_pad_shape([[0, 0], [1, 0], [0, 0]]))[:, :-1]
+    path = path.unsqueeze(1).transpose(2, 3) * mask
+    return path
+
+
+def clip_grad_value_(parameters, clip_value, norm_type=2):
+    if isinstance(parameters, torch.Tensor):
+        parameters = [parameters]
+    parameters = list(filter(lambda p: p.grad is not None, parameters))
+    norm_type = float(norm_type)
+    if clip_value is not None:
+        clip_value = float(clip_value)
+
+    total_norm = 0
+    for p in parameters:
+        param_norm = p.grad.data.norm(norm_type)
+        total_norm += param_norm.item() ** norm_type
+        if clip_value is not None:
+            p.grad.data.clamp_(min=-clip_value, max=clip_value)
+    total_norm = total_norm ** (1.0 / norm_type)
+    return total_norm

lib/infer_pack/models.py

@@ -0,0 +1,1142 @@
+import math, pdb, os
+from time import time as ttime
+import torch
+from torch import nn
+from torch.nn import functional as F
+from lib.infer_pack import modules
+from lib.infer_pack import attentions
+from lib.infer_pack import commons
+from lib.infer_pack.commons import init_weights, get_padding
+from torch.nn import Conv1d, ConvTranspose1d, AvgPool1d, Conv2d
+from torch.nn.utils import weight_norm, remove_weight_norm, spectral_norm
+from lib.infer_pack.commons import init_weights
+import numpy as np
+from lib.infer_pack import commons
+
+
+class TextEncoder256(nn.Module):
+    def __init__(
+        self,
+        out_channels,
+        hidden_channels,
+        filter_channels,
+        n_heads,
+        n_layers,
+        kernel_size,
+        p_dropout,
+        f0=True,
+    ):
+        super().__init__()
+        self.out_channels = out_channels
+        self.hidden_channels = hidden_channels
+        self.filter_channels = filter_channels
+        self.n_heads = n_heads
+        self.n_layers = n_layers
+        self.kernel_size = kernel_size
+        self.p_dropout = p_dropout
+        self.emb_phone = nn.Linear(256, hidden_channels)
+        self.lrelu = nn.LeakyReLU(0.1, inplace=True)
+        if f0 == True:
+            self.emb_pitch = nn.Embedding(256, hidden_channels)  # pitch 256
+        self.encoder = attentions.Encoder(
+            hidden_channels, filter_channels, n_heads, n_layers, kernel_size, p_dropout
+        )
+        self.proj = nn.Conv1d(hidden_channels, out_channels * 2, 1)
+
+    def forward(self, phone, pitch, lengths):
+        if pitch == None:
+            x = self.emb_phone(phone)
+        else:
+            x = self.emb_phone(phone) + self.emb_pitch(pitch)
+        x = x * math.sqrt(self.hidden_channels)  # [b, t, h]
+        x = self.lrelu(x)
+        x = torch.transpose(x, 1, -1)  # [b, h, t]
+        x_mask = torch.unsqueeze(commons.sequence_mask(lengths, x.size(2)), 1).to(
+            x.dtype
+        )
+        x = self.encoder(x * x_mask, x_mask)
+        stats = self.proj(x) * x_mask
+
+        m, logs = torch.split(stats, self.out_channels, dim=1)
+        return m, logs, x_mask
+
+
+class TextEncoder768(nn.Module):
+    def __init__(
+        self,
+        out_channels,
+        hidden_channels,
+        filter_channels,
+        n_heads,
+        n_layers,
+        kernel_size,
+        p_dropout,
+        f0=True,
+    ):
+        super().__init__()
+        self.out_channels = out_channels
+        self.hidden_channels = hidden_channels
+        self.filter_channels = filter_channels
+        self.n_heads = n_heads
+        self.n_layers = n_layers
+        self.kernel_size = kernel_size
+        self.p_dropout = p_dropout
+        self.emb_phone = nn.Linear(768, hidden_channels)
+        self.lrelu = nn.LeakyReLU(0.1, inplace=True)
+        if f0 == True:
+            self.emb_pitch = nn.Embedding(256, hidden_channels)  # pitch 256
+        self.encoder = attentions.Encoder(
+            hidden_channels, filter_channels, n_heads, n_layers, kernel_size, p_dropout
+        )
+        self.proj = nn.Conv1d(hidden_channels, out_channels * 2, 1)
+
+    def forward(self, phone, pitch, lengths):
+        if pitch == None:
+            x = self.emb_phone(phone)
+        else:
+            x = self.emb_phone(phone) + self.emb_pitch(pitch)
+        x = x * math.sqrt(self.hidden_channels)  # [b, t, h]
+        x = self.lrelu(x)
+        x = torch.transpose(x, 1, -1)  # [b, h, t]
+        x_mask = torch.unsqueeze(commons.sequence_mask(lengths, x.size(2)), 1).to(
+            x.dtype
+        )
+        x = self.encoder(x * x_mask, x_mask)
+        stats = self.proj(x) * x_mask
+
+        m, logs = torch.split(stats, self.out_channels, dim=1)
+        return m, logs, x_mask
+
+
+class ResidualCouplingBlock(nn.Module):
+    def __init__(
+        self,
+        channels,
+        hidden_channels,
+        kernel_size,
+        dilation_rate,
+        n_layers,
+        n_flows=4,
+        gin_channels=0,
+    ):
+        super().__init__()
+        self.channels = channels
+        self.hidden_channels = hidden_channels
+        self.kernel_size = kernel_size
+        self.dilation_rate = dilation_rate
+        self.n_layers = n_layers
+        self.n_flows = n_flows
+        self.gin_channels = gin_channels
+
+        self.flows = nn.ModuleList()
+        for i in range(n_flows):
+            self.flows.append(
+                modules.ResidualCouplingLayer(
+                    channels,
+                    hidden_channels,
+                    kernel_size,
+                    dilation_rate,
+                    n_layers,
+                    gin_channels=gin_channels,
+                    mean_only=True,
+                )
+            )
+            self.flows.append(modules.Flip())
+
+    def forward(self, x, x_mask, g=None, reverse=False):
+        if not reverse:
+            for flow in self.flows:
+                x, _ = flow(x, x_mask, g=g, reverse=reverse)
+        else:
+            for flow in reversed(self.flows):
+                x = flow(x, x_mask, g=g, reverse=reverse)
+        return x
+
+    def remove_weight_norm(self):
+        for i in range(self.n_flows):
+            self.flows[i * 2].remove_weight_norm()
+
+
+class PosteriorEncoder(nn.Module):
+    def __init__(
+        self,
+        in_channels,
+        out_channels,
+        hidden_channels,
+        kernel_size,
+        dilation_rate,
+        n_layers,
+        gin_channels=0,
+    ):
+        super().__init__()
+        self.in_channels = in_channels
+        self.out_channels = out_channels
+        self.hidden_channels = hidden_channels
+        self.kernel_size = kernel_size
+        self.dilation_rate = dilation_rate
+        self.n_layers = n_layers
+        self.gin_channels = gin_channels
+
+        self.pre = nn.Conv1d(in_channels, hidden_channels, 1)
+        self.enc = modules.WN(
+            hidden_channels,
+            kernel_size,
+            dilation_rate,
+            n_layers,
+            gin_channels=gin_channels,
+        )
+        self.proj = nn.Conv1d(hidden_channels, out_channels * 2, 1)
+
+    def forward(self, x, x_lengths, g=None):
+        x_mask = torch.unsqueeze(commons.sequence_mask(x_lengths, x.size(2)), 1).to(
+            x.dtype
+        )
+        x = self.pre(x) * x_mask
+        x = self.enc(x, x_mask, g=g)
+        stats = self.proj(x) * x_mask
+        m, logs = torch.split(stats, self.out_channels, dim=1)
+        z = (m + torch.randn_like(m) * torch.exp(logs)) * x_mask
+        return z, m, logs, x_mask
+
+    def remove_weight_norm(self):
+        self.enc.remove_weight_norm()
+
+
+class Generator(torch.nn.Module):
+    def __init__(
+        self,
+        initial_channel,
+        resblock,
+        resblock_kernel_sizes,
+        resblock_dilation_sizes,
+        upsample_rates,
+        upsample_initial_channel,
+        upsample_kernel_sizes,
+        gin_channels=0,
+    ):
+        super(Generator, self).__init__()
+        self.num_kernels = len(resblock_kernel_sizes)
+        self.num_upsamples = len(upsample_rates)
+        self.conv_pre = Conv1d(
+            initial_channel, upsample_initial_channel, 7, 1, padding=3
+        )
+        resblock = modules.ResBlock1 if resblock == "1" else modules.ResBlock2
+
+        self.ups = nn.ModuleList()
+        for i, (u, k) in enumerate(zip(upsample_rates, upsample_kernel_sizes)):
+            self.ups.append(
+                weight_norm(
+                    ConvTranspose1d(
+                        upsample_initial_channel // (2**i),
+                        upsample_initial_channel // (2 ** (i + 1)),
+                        k,
+                        u,
+                        padding=(k - u) // 2,
+                    )
+                )
+            )
+
+        self.resblocks = nn.ModuleList()
+        for i in range(len(self.ups)):
+            ch = upsample_initial_channel // (2 ** (i + 1))
+            for j, (k, d) in enumerate(
+                zip(resblock_kernel_sizes, resblock_dilation_sizes)
+            ):
+                self.resblocks.append(resblock(ch, k, d))
+
+        self.conv_post = Conv1d(ch, 1, 7, 1, padding=3, bias=False)
+        self.ups.apply(init_weights)
+
+        if gin_channels != 0:
+            self.cond = nn.Conv1d(gin_channels, upsample_initial_channel, 1)
+
+    def forward(self, x, g=None):
+        x = self.conv_pre(x)
+        if g is not None:
+            x = x + self.cond(g)
+
+        for i in range(self.num_upsamples):
+            x = F.leaky_relu(x, modules.LRELU_SLOPE)
+            x = self.ups[i](x)
+            xs = None
+            for j in range(self.num_kernels):
+                if xs is None:
+                    xs = self.resblocks[i * self.num_kernels + j](x)
+                else:
+                    xs += self.resblocks[i * self.num_kernels + j](x)
+            x = xs / self.num_kernels
+        x = F.leaky_relu(x)
+        x = self.conv_post(x)
+        x = torch.tanh(x)
+
+        return x
+
+    def remove_weight_norm(self):
+        for l in self.ups:
+            remove_weight_norm(l)
+        for l in self.resblocks:
+            l.remove_weight_norm()
+
+
+class SineGen(torch.nn.Module):
+    """Definition of sine generator
+    SineGen(samp_rate, harmonic_num = 0,
+            sine_amp = 0.1, noise_std = 0.003,
+            voiced_threshold = 0,
+            flag_for_pulse=False)
+    samp_rate: sampling rate in Hz
+    harmonic_num: number of harmonic overtones (default 0)
+    sine_amp: amplitude of sine-wavefrom (default 0.1)
+    noise_std: std of Gaussian noise (default 0.003)
+    voiced_thoreshold: F0 threshold for U/V classification (default 0)
+    flag_for_pulse: this SinGen is used inside PulseGen (default False)
+    Note: when flag_for_pulse is True, the first time step of a voiced
+        segment is always sin(np.pi) or cos(0)
+    """
+
+    def __init__(
+        self,
+        samp_rate,
+        harmonic_num=0,
+        sine_amp=0.1,
+        noise_std=0.003,
+        voiced_threshold=0,
+        flag_for_pulse=False,
+    ):
+        super(SineGen, self).__init__()
+        self.sine_amp = sine_amp
+        self.noise_std = noise_std
+        self.harmonic_num = harmonic_num
+        self.dim = self.harmonic_num + 1
+        self.sampling_rate = samp_rate
+        self.voiced_threshold = voiced_threshold
+
+    def _f02uv(self, f0):
+        # generate uv signal
+        uv = torch.ones_like(f0)
+        uv = uv * (f0 > self.voiced_threshold)
+        return uv
+
+    def forward(self, f0, upp):
+        """sine_tensor, uv = forward(f0)
+        input F0: tensor(batchsize=1, length, dim=1)
+                  f0 for unvoiced steps should be 0
+        output sine_tensor: tensor(batchsize=1, length, dim)
+        output uv: tensor(batchsize=1, length, 1)
+        """
+        with torch.no_grad():
+            f0 = f0[:, None].transpose(1, 2)
+            f0_buf = torch.zeros(f0.shape[0], f0.shape[1], self.dim, device=f0.device)
+            # fundamental component
+            f0_buf[:, :, 0] = f0[:, :, 0]
+            for idx in np.arange(self.harmonic_num):
+                f0_buf[:, :, idx + 1] = f0_buf[:, :, 0] * (
+                    idx + 2
+                )  # idx + 2: the (idx+1)-th overtone, (idx+2)-th harmonic
+            rad_values = (f0_buf / self.sampling_rate) % 1  ###%1 means that the product of n_har cannot be post-processed and optimized
+            rand_ini = torch.rand(
+                f0_buf.shape[0], f0_buf.shape[2], device=f0_buf.device
+            )
+            rand_ini[:, 0] = 0
+            rad_values[:, 0, :] = rad_values[:, 0, :] + rand_ini
+            tmp_over_one = torch.cumsum(rad_values, 1)  # % 1  #####%1 means that the following cumsum can no longer be optimized
+            tmp_over_one *= upp
+            tmp_over_one = F.interpolate(
+                tmp_over_one.transpose(2, 1),
+                scale_factor=upp,
+                mode="linear",
+                align_corners=True,
+            ).transpose(2, 1)
+            rad_values = F.interpolate(
+                rad_values.transpose(2, 1), scale_factor=upp, mode="nearest"
+            ).transpose(
+                2, 1
+            )  #######
+            tmp_over_one %= 1
+            tmp_over_one_idx = (tmp_over_one[:, 1:, :] - tmp_over_one[:, :-1, :]) < 0
+            cumsum_shift = torch.zeros_like(rad_values)
+            cumsum_shift[:, 1:, :] = tmp_over_one_idx * -1.0
+            sine_waves = torch.sin(
+                torch.cumsum(rad_values + cumsum_shift, dim=1) * 2 * np.pi
+            )
+            sine_waves = sine_waves * self.sine_amp
+            uv = self._f02uv(f0)
+            uv = F.interpolate(
+                uv.transpose(2, 1), scale_factor=upp, mode="nearest"
+            ).transpose(2, 1)
+            noise_amp = uv * self.noise_std + (1 - uv) * self.sine_amp / 3
+            noise = noise_amp * torch.randn_like(sine_waves)
+            sine_waves = sine_waves * uv + noise
+        return sine_waves, uv, noise
+
+
+class SourceModuleHnNSF(torch.nn.Module):
+    """SourceModule for hn-nsf
+    SourceModule(sampling_rate, harmonic_num=0, sine_amp=0.1,
+                 add_noise_std=0.003, voiced_threshod=0)
+    sampling_rate: sampling_rate in Hz
+    harmonic_num: number of harmonic above F0 (default: 0)
+    sine_amp: amplitude of sine source signal (default: 0.1)
+    add_noise_std: std of additive Gaussian noise (default: 0.003)
+        note that amplitude of noise in unvoiced is decided
+        by sine_amp
+    voiced_threshold: threhold to set U/V given F0 (default: 0)
+    Sine_source, noise_source = SourceModuleHnNSF(F0_sampled)
+    F0_sampled (batchsize, length, 1)
+    Sine_source (batchsize, length, 1)
+    noise_source (batchsize, length 1)
+    uv (batchsize, length, 1)
+    """
+
+    def __init__(
+        self,
+        sampling_rate,
+        harmonic_num=0,
+        sine_amp=0.1,
+        add_noise_std=0.003,
+        voiced_threshod=0,
+        is_half=True,
+    ):
+        super(SourceModuleHnNSF, self).__init__()
+
+        self.sine_amp = sine_amp
+        self.noise_std = add_noise_std
+        self.is_half = is_half
+        # to produce sine waveforms
+        self.l_sin_gen = SineGen(
+            sampling_rate, harmonic_num, sine_amp, add_noise_std, voiced_threshod
+        )
+
+        # to merge source harmonics into a single excitation
+        self.l_linear = torch.nn.Linear(harmonic_num + 1, 1)
+        self.l_tanh = torch.nn.Tanh()
+
+    def forward(self, x, upp=None):
+        sine_wavs, uv, _ = self.l_sin_gen(x, upp)
+        if self.is_half:
+            sine_wavs = sine_wavs.half()
+        sine_merge = self.l_tanh(self.l_linear(sine_wavs))
+        return sine_merge, None, None  # noise, uv
+
+
+class GeneratorNSF(torch.nn.Module):
+    def __init__(
+        self,
+        initial_channel,
+        resblock,
+        resblock_kernel_sizes,
+        resblock_dilation_sizes,
+        upsample_rates,
+        upsample_initial_channel,
+        upsample_kernel_sizes,
+        gin_channels,
+        sr,
+        is_half=False,
+    ):
+        super(GeneratorNSF, self).__init__()
+        self.num_kernels = len(resblock_kernel_sizes)
+        self.num_upsamples = len(upsample_rates)
+
+        self.f0_upsamp = torch.nn.Upsample(scale_factor=np.prod(upsample_rates))
+        self.m_source = SourceModuleHnNSF(
+            sampling_rate=sr, harmonic_num=0, is_half=is_half
+        )
+        self.noise_convs = nn.ModuleList()
+        self.conv_pre = Conv1d(
+            initial_channel, upsample_initial_channel, 7, 1, padding=3
+        )
+        resblock = modules.ResBlock1 if resblock == "1" else modules.ResBlock2
+
+        self.ups = nn.ModuleList()
+        for i, (u, k) in enumerate(zip(upsample_rates, upsample_kernel_sizes)):
+            c_cur = upsample_initial_channel // (2 ** (i + 1))
+            self.ups.append(
+                weight_norm(
+                    ConvTranspose1d(
+                        upsample_initial_channel // (2**i),
+                        upsample_initial_channel // (2 ** (i + 1)),
+                        k,
+                        u,
+                        padding=(k - u) // 2,
+                    )
+                )
+            )
+            if i + 1 < len(upsample_rates):
+                stride_f0 = np.prod(upsample_rates[i + 1 :])
+                self.noise_convs.append(
+                    Conv1d(
+                        1,
+                        c_cur,
+                        kernel_size=stride_f0 * 2,
+                        stride=stride_f0,
+                        padding=stride_f0 // 2,
+                    )
+                )
+            else:
+                self.noise_convs.append(Conv1d(1, c_cur, kernel_size=1))
+
+        self.resblocks = nn.ModuleList()
+        for i in range(len(self.ups)):
+            ch = upsample_initial_channel // (2 ** (i + 1))
+            for j, (k, d) in enumerate(
+                zip(resblock_kernel_sizes, resblock_dilation_sizes)
+            ):
+                self.resblocks.append(resblock(ch, k, d))
+
+        self.conv_post = Conv1d(ch, 1, 7, 1, padding=3, bias=False)
+        self.ups.apply(init_weights)
+
+        if gin_channels != 0:
+            self.cond = nn.Conv1d(gin_channels, upsample_initial_channel, 1)
+
+        self.upp = np.prod(upsample_rates)
+
+    def forward(self, x, f0, g=None):
+        har_source, noi_source, uv = self.m_source(f0, self.upp)
+        har_source = har_source.transpose(1, 2)
+        x = self.conv_pre(x)
+        if g is not None:
+            x = x + self.cond(g)
+
+        for i in range(self.num_upsamples):
+            x = F.leaky_relu(x, modules.LRELU_SLOPE)
+            x = self.ups[i](x)
+            x_source = self.noise_convs[i](har_source)
+            x = x + x_source
+            xs = None
+            for j in range(self.num_kernels):
+                if xs is None:
+                    xs = self.resblocks[i * self.num_kernels + j](x)
+                else:
+                    xs += self.resblocks[i * self.num_kernels + j](x)
+            x = xs / self.num_kernels
+        x = F.leaky_relu(x)
+        x = self.conv_post(x)
+        x = torch.tanh(x)
+        return x
+
+    def remove_weight_norm(self):
+        for l in self.ups:
+            remove_weight_norm(l)
+        for l in self.resblocks:
+            l.remove_weight_norm()
+
+
+sr2sr = {
+    "32k": 32000,
+    "40k": 40000,
+    "48k": 48000,
+}
+
+
+class SynthesizerTrnMs256NSFsid(nn.Module):
+    def __init__(
+        self,
+        spec_channels,
+        segment_size,
+        inter_channels,
+        hidden_channels,
+        filter_channels,
+        n_heads,
+        n_layers,
+        kernel_size,
+        p_dropout,
+        resblock,
+        resblock_kernel_sizes,
+        resblock_dilation_sizes,
+        upsample_rates,
+        upsample_initial_channel,
+        upsample_kernel_sizes,
+        spk_embed_dim,
+        gin_channels,
+        sr,
+        **kwargs
+    ):
+        super().__init__()
+        if type(sr) == type("strr"):
+            sr = sr2sr[sr]
+        self.spec_channels = spec_channels
+        self.inter_channels = inter_channels
+        self.hidden_channels = hidden_channels
+        self.filter_channels = filter_channels
+        self.n_heads = n_heads
+        self.n_layers = n_layers
+        self.kernel_size = kernel_size
+        self.p_dropout = p_dropout
+        self.resblock = resblock
+        self.resblock_kernel_sizes = resblock_kernel_sizes
+        self.resblock_dilation_sizes = resblock_dilation_sizes
+        self.upsample_rates = upsample_rates
+        self.upsample_initial_channel = upsample_initial_channel
+        self.upsample_kernel_sizes = upsample_kernel_sizes
+        self.segment_size = segment_size
+        self.gin_channels = gin_channels
+        # self.hop_length = hop_length#
+        self.spk_embed_dim = spk_embed_dim
+        self.enc_p = TextEncoder256(
+            inter_channels,
+            hidden_channels,
+            filter_channels,
+            n_heads,
+            n_layers,
+            kernel_size,
+            p_dropout,
+        )
+        self.dec = GeneratorNSF(
+            inter_channels,
+            resblock,
+            resblock_kernel_sizes,
+            resblock_dilation_sizes,
+            upsample_rates,
+            upsample_initial_channel,
+            upsample_kernel_sizes,
+            gin_channels=gin_channels,
+            sr=sr,
+            is_half=kwargs["is_half"],
+        )
+        self.enc_q = PosteriorEncoder(
+            spec_channels,
+            inter_channels,
+            hidden_channels,
+            5,
+            1,
+            16,
+            gin_channels=gin_channels,
+        )
+        self.flow = ResidualCouplingBlock(
+            inter_channels, hidden_channels, 5, 1, 3, gin_channels=gin_channels
+        )
+        self.emb_g = nn.Embedding(self.spk_embed_dim, gin_channels)
+        print("gin_channels:", gin_channels, "self.spk_embed_dim:", self.spk_embed_dim)
+
+    def remove_weight_norm(self):
+        self.dec.remove_weight_norm()
+        self.flow.remove_weight_norm()
+        self.enc_q.remove_weight_norm()
+
+    def forward(
+        self, phone, phone_lengths, pitch, pitchf, y, y_lengths, ds
+    ):  # Here ds is id, [bs,1]
+        # print(1,pitch.shape)#[bs,t]
+        g = self.emb_g(ds).unsqueeze(-1)  # [b, 256, 1]##1 is t, broadcast
+        m_p, logs_p, x_mask = self.enc_p(phone, pitch, phone_lengths)
+        z, m_q, logs_q, y_mask = self.enc_q(y, y_lengths, g=g)
+        z_p = self.flow(z, y_mask, g=g)
+        z_slice, ids_slice = commons.rand_slice_segments(
+            z, y_lengths, self.segment_size
+        )
+        # print(-1,pitchf.shape,ids_slice,self.segment_size,self.hop_length,self.segment_size//self.hop_length)
+        pitchf = commons.slice_segments2(pitchf, ids_slice, self.segment_size)
+        # print(-2,pitchf.shape,z_slice.shape)
+        o = self.dec(z_slice, pitchf, g=g)
+        return o, ids_slice, x_mask, y_mask, (z, z_p, m_p, logs_p, m_q, logs_q)
+
+    def infer(self, phone, phone_lengths, pitch, nsff0, sid, rate=None):
+        g = self.emb_g(sid).unsqueeze(-1)
+        m_p, logs_p, x_mask = self.enc_p(phone, pitch, phone_lengths)
+        z_p = (m_p + torch.exp(logs_p) * torch.randn_like(m_p) * 0.66666) * x_mask
+        if rate:
+            head = int(z_p.shape[2] * rate)
+            z_p = z_p[:, :, -head:]
+            x_mask = x_mask[:, :, -head:]
+            nsff0 = nsff0[:, -head:]
+        z = self.flow(z_p, x_mask, g=g, reverse=True)
+        o = self.dec(z * x_mask, nsff0, g=g)
+        return o, x_mask, (z, z_p, m_p, logs_p)
+
+
+class SynthesizerTrnMs768NSFsid(nn.Module):
+    def __init__(
+        self,
+        spec_channels,
+        segment_size,
+        inter_channels,
+        hidden_channels,
+        filter_channels,
+        n_heads,
+        n_layers,
+        kernel_size,
+        p_dropout,
+        resblock,
+        resblock_kernel_sizes,
+        resblock_dilation_sizes,
+        upsample_rates,
+        upsample_initial_channel,
+        upsample_kernel_sizes,
+        spk_embed_dim,
+        gin_channels,
+        sr,
+        **kwargs
+    ):
+        super().__init__()
+        if type(sr) == type("strr"):
+            sr = sr2sr[sr]
+        self.spec_channels = spec_channels
+        self.inter_channels = inter_channels
+        self.hidden_channels = hidden_channels
+        self.filter_channels = filter_channels
+        self.n_heads = n_heads
+        self.n_layers = n_layers
+        self.kernel_size = kernel_size
+        self.p_dropout = p_dropout
+        self.resblock = resblock
+        self.resblock_kernel_sizes = resblock_kernel_sizes
+        self.resblock_dilation_sizes = resblock_dilation_sizes
+        self.upsample_rates = upsample_rates
+        self.upsample_initial_channel = upsample_initial_channel
+        self.upsample_kernel_sizes = upsample_kernel_sizes
+        self.segment_size = segment_size
+        self.gin_channels = gin_channels
+        # self.hop_length = hop_length#
+        self.spk_embed_dim = spk_embed_dim
+        self.enc_p = TextEncoder768(
+            inter_channels,
+            hidden_channels,
+            filter_channels,
+            n_heads,
+            n_layers,
+            kernel_size,
+            p_dropout,
+        )
+        self.dec = GeneratorNSF(
+            inter_channels,
+            resblock,
+            resblock_kernel_sizes,
+            resblock_dilation_sizes,
+            upsample_rates,
+            upsample_initial_channel,
+            upsample_kernel_sizes,
+            gin_channels=gin_channels,
+            sr=sr,
+            is_half=kwargs["is_half"],
+        )
+        self.enc_q = PosteriorEncoder(
+            spec_channels,
+            inter_channels,
+            hidden_channels,
+            5,
+            1,
+            16,
+            gin_channels=gin_channels,
+        )
+        self.flow = ResidualCouplingBlock(
+            inter_channels, hidden_channels, 5, 1, 3, gin_channels=gin_channels
+        )
+        self.emb_g = nn.Embedding(self.spk_embed_dim, gin_channels)
+        print("gin_channels:", gin_channels, "self.spk_embed_dim:", self.spk_embed_dim)
+
+    def remove_weight_norm(self):
+        self.dec.remove_weight_norm()
+        self.flow.remove_weight_norm()
+        self.enc_q.remove_weight_norm()
+
+    def forward(
+        self, phone, phone_lengths, pitch, pitchf, y, y_lengths, ds
+    ):  # Here ds is id,[bs,1]
+        # print(1,pitch.shape)#[bs,t]
+        g = self.emb_g(ds).unsqueeze(-1)  # [b, 256, 1]##1 is t, broadcast
+        m_p, logs_p, x_mask = self.enc_p(phone, pitch, phone_lengths)
+        z, m_q, logs_q, y_mask = self.enc_q(y, y_lengths, g=g)
+        z_p = self.flow(z, y_mask, g=g)
+        z_slice, ids_slice = commons.rand_slice_segments(
+            z, y_lengths, self.segment_size
+        )
+        # print(-1,pitchf.shape,ids_slice,self.segment_size,self.hop_length,self.segment_size//self.hop_length)
+        pitchf = commons.slice_segments2(pitchf, ids_slice, self.segment_size)
+        # print(-2,pitchf.shape,z_slice.shape)
+        o = self.dec(z_slice, pitchf, g=g)
+        return o, ids_slice, x_mask, y_mask, (z, z_p, m_p, logs_p, m_q, logs_q)
+
+    def infer(self, phone, phone_lengths, pitch, nsff0, sid, rate=None):
+        g = self.emb_g(sid).unsqueeze(-1)
+        m_p, logs_p, x_mask = self.enc_p(phone, pitch, phone_lengths)
+        z_p = (m_p + torch.exp(logs_p) * torch.randn_like(m_p) * 0.66666) * x_mask
+        if rate:
+            head = int(z_p.shape[2] * rate)
+            z_p = z_p[:, :, -head:]
+            x_mask = x_mask[:, :, -head:]
+            nsff0 = nsff0[:, -head:]
+        z = self.flow(z_p, x_mask, g=g, reverse=True)
+        o = self.dec(z * x_mask, nsff0, g=g)
+        return o, x_mask, (z, z_p, m_p, logs_p)
+
+
+class SynthesizerTrnMs256NSFsid_nono(nn.Module):
+    def __init__(
+        self,
+        spec_channels,
+        segment_size,
+        inter_channels,
+        hidden_channels,
+        filter_channels,
+        n_heads,
+        n_layers,
+        kernel_size,
+        p_dropout,
+        resblock,
+        resblock_kernel_sizes,
+        resblock_dilation_sizes,
+        upsample_rates,
+        upsample_initial_channel,
+        upsample_kernel_sizes,
+        spk_embed_dim,
+        gin_channels,
+        sr=None,
+        **kwargs
+    ):
+        super().__init__()
+        self.spec_channels = spec_channels
+        self.inter_channels = inter_channels
+        self.hidden_channels = hidden_channels
+        self.filter_channels = filter_channels
+        self.n_heads = n_heads
+        self.n_layers = n_layers
+        self.kernel_size = kernel_size
+        self.p_dropout = p_dropout
+        self.resblock = resblock
+        self.resblock_kernel_sizes = resblock_kernel_sizes
+        self.resblock_dilation_sizes = resblock_dilation_sizes
+        self.upsample_rates = upsample_rates
+        self.upsample_initial_channel = upsample_initial_channel
+        self.upsample_kernel_sizes = upsample_kernel_sizes
+        self.segment_size = segment_size
+        self.gin_channels = gin_channels
+        # self.hop_length = hop_length#
+        self.spk_embed_dim = spk_embed_dim
+        self.enc_p = TextEncoder256(
+            inter_channels,
+            hidden_channels,
+            filter_channels,
+            n_heads,
+            n_layers,
+            kernel_size,
+            p_dropout,
+            f0=False,
+        )
+        self.dec = Generator(
+            inter_channels,
+            resblock,
+            resblock_kernel_sizes,
+            resblock_dilation_sizes,
+            upsample_rates,
+            upsample_initial_channel,
+            upsample_kernel_sizes,
+            gin_channels=gin_channels,
+        )
+        self.enc_q = PosteriorEncoder(
+            spec_channels,
+            inter_channels,
+            hidden_channels,
+            5,
+            1,
+            16,
+            gin_channels=gin_channels,
+        )
+        self.flow = ResidualCouplingBlock(
+            inter_channels, hidden_channels, 5, 1, 3, gin_channels=gin_channels
+        )
+        self.emb_g = nn.Embedding(self.spk_embed_dim, gin_channels)
+        print("gin_channels:", gin_channels, "self.spk_embed_dim:", self.spk_embed_dim)
+
+    def remove_weight_norm(self):
+        self.dec.remove_weight_norm()
+        self.flow.remove_weight_norm()
+        self.enc_q.remove_weight_norm()
+
+    def forward(self, phone, phone_lengths, y, y_lengths, ds):  # Here ds is id,[bs,1]
+        g = self.emb_g(ds).unsqueeze(-1)  # [b, 256, 1]##1 is t, broadcast
+        m_p, logs_p, x_mask = self.enc_p(phone, None, phone_lengths)
+        z, m_q, logs_q, y_mask = self.enc_q(y, y_lengths, g=g)
+        z_p = self.flow(z, y_mask, g=g)
+        z_slice, ids_slice = commons.rand_slice_segments(
+            z, y_lengths, self.segment_size
+        )
+        o = self.dec(z_slice, g=g)
+        return o, ids_slice, x_mask, y_mask, (z, z_p, m_p, logs_p, m_q, logs_q)
+
+    def infer(self, phone, phone_lengths, sid, rate=None):
+        g = self.emb_g(sid).unsqueeze(-1)
+        m_p, logs_p, x_mask = self.enc_p(phone, None, phone_lengths)
+        z_p = (m_p + torch.exp(logs_p) * torch.randn_like(m_p) * 0.66666) * x_mask
+        if rate:
+            head = int(z_p.shape[2] * rate)
+            z_p = z_p[:, :, -head:]
+            x_mask = x_mask[:, :, -head:]
+        z = self.flow(z_p, x_mask, g=g, reverse=True)
+        o = self.dec(z * x_mask, g=g)
+        return o, x_mask, (z, z_p, m_p, logs_p)
+
+
+class SynthesizerTrnMs768NSFsid_nono(nn.Module):
+    def __init__(
+        self,
+        spec_channels,
+        segment_size,
+        inter_channels,
+        hidden_channels,
+        filter_channels,
+        n_heads,
+        n_layers,
+        kernel_size,
+        p_dropout,
+        resblock,
+        resblock_kernel_sizes,
+        resblock_dilation_sizes,
+        upsample_rates,
+        upsample_initial_channel,
+        upsample_kernel_sizes,
+        spk_embed_dim,
+        gin_channels,
+        sr=None,
+        **kwargs
+    ):
+        super().__init__()
+        self.spec_channels = spec_channels
+        self.inter_channels = inter_channels
+        self.hidden_channels = hidden_channels
+        self.filter_channels = filter_channels
+        self.n_heads = n_heads
+        self.n_layers = n_layers
+        self.kernel_size = kernel_size
+        self.p_dropout = p_dropout
+        self.resblock = resblock
+        self.resblock_kernel_sizes = resblock_kernel_sizes
+        self.resblock_dilation_sizes = resblock_dilation_sizes
+        self.upsample_rates = upsample_rates
+        self.upsample_initial_channel = upsample_initial_channel
+        self.upsample_kernel_sizes = upsample_kernel_sizes
+        self.segment_size = segment_size
+        self.gin_channels = gin_channels
+        # self.hop_length = hop_length#
+        self.spk_embed_dim = spk_embed_dim
+        self.enc_p = TextEncoder768(
+            inter_channels,
+            hidden_channels,
+            filter_channels,
+            n_heads,
+            n_layers,
+            kernel_size,
+            p_dropout,
+            f0=False,
+        )
+        self.dec = Generator(
+            inter_channels,
+            resblock,
+            resblock_kernel_sizes,
+            resblock_dilation_sizes,
+            upsample_rates,
+            upsample_initial_channel,
+            upsample_kernel_sizes,
+            gin_channels=gin_channels,
+        )
+        self.enc_q = PosteriorEncoder(
+            spec_channels,
+            inter_channels,
+            hidden_channels,
+            5,
+            1,
+            16,
+            gin_channels=gin_channels,
+        )
+        self.flow = ResidualCouplingBlock(
+            inter_channels, hidden_channels, 5, 1, 3, gin_channels=gin_channels
+        )
+        self.emb_g = nn.Embedding(self.spk_embed_dim, gin_channels)
+        print("gin_channels:", gin_channels, "self.spk_embed_dim:", self.spk_embed_dim)
+
+    def remove_weight_norm(self):
+        self.dec.remove_weight_norm()
+        self.flow.remove_weight_norm()
+        self.enc_q.remove_weight_norm()
+
+    def forward(self, phone, phone_lengths, y, y_lengths, ds):  # Here ds is id,[bs,1]
+        g = self.emb_g(ds).unsqueeze(-1)  # [b, 256, 1]##1 is t, broadcast
+        m_p, logs_p, x_mask = self.enc_p(phone, None, phone_lengths)
+        z, m_q, logs_q, y_mask = self.enc_q(y, y_lengths, g=g)
+        z_p = self.flow(z, y_mask, g=g)
+        z_slice, ids_slice = commons.rand_slice_segments(
+            z, y_lengths, self.segment_size
+        )
+        o = self.dec(z_slice, g=g)
+        return o, ids_slice, x_mask, y_mask, (z, z_p, m_p, logs_p, m_q, logs_q)
+
+    def infer(self, phone, phone_lengths, sid, rate=None):
+        g = self.emb_g(sid).unsqueeze(-1)
+        m_p, logs_p, x_mask = self.enc_p(phone, None, phone_lengths)
+        z_p = (m_p + torch.exp(logs_p) * torch.randn_like(m_p) * 0.66666) * x_mask
+        if rate:
+            head = int(z_p.shape[2] * rate)
+            z_p = z_p[:, :, -head:]
+            x_mask = x_mask[:, :, -head:]
+        z = self.flow(z_p, x_mask, g=g, reverse=True)
+        o = self.dec(z * x_mask, g=g)
+        return o, x_mask, (z, z_p, m_p, logs_p)
+
+
+class MultiPeriodDiscriminator(torch.nn.Module):
+    def __init__(self, use_spectral_norm=False):
+        super(MultiPeriodDiscriminator, self).__init__()
+        periods = [2, 3, 5, 7, 11, 17]
+        # periods = [3, 5, 7, 11, 17, 23, 37]
+
+        discs = [DiscriminatorS(use_spectral_norm=use_spectral_norm)]
+        discs = discs + [
+            DiscriminatorP(i, use_spectral_norm=use_spectral_norm) for i in periods
+        ]
+        self.discriminators = nn.ModuleList(discs)
+
+    def forward(self, y, y_hat):
+        y_d_rs = []  #
+        y_d_gs = []
+        fmap_rs = []
+        fmap_gs = []
+        for i, d in enumerate(self.discriminators):
+            y_d_r, fmap_r = d(y)
+            y_d_g, fmap_g = d(y_hat)
+            # for j in range(len(fmap_r)):
+            #     print(i,j,y.shape,y_hat.shape,fmap_r[j].shape,fmap_g[j].shape)
+            y_d_rs.append(y_d_r)
+            y_d_gs.append(y_d_g)
+            fmap_rs.append(fmap_r)
+            fmap_gs.append(fmap_g)
+
+        return y_d_rs, y_d_gs, fmap_rs, fmap_gs
+
+
+class MultiPeriodDiscriminatorV2(torch.nn.Module):
+    def __init__(self, use_spectral_norm=False):
+        super(MultiPeriodDiscriminatorV2, self).__init__()
+        # periods = [2, 3, 5, 7, 11, 17]
+        periods = [2, 3, 5, 7, 11, 17, 23, 37]
+
+        discs = [DiscriminatorS(use_spectral_norm=use_spectral_norm)]
+        discs = discs + [
+            DiscriminatorP(i, use_spectral_norm=use_spectral_norm) for i in periods
+        ]
+        self.discriminators = nn.ModuleList(discs)
+
+    def forward(self, y, y_hat):
+        y_d_rs = []  #
+        y_d_gs = []
+        fmap_rs = []
+        fmap_gs = []
+        for i, d in enumerate(self.discriminators):
+            y_d_r, fmap_r = d(y)
+            y_d_g, fmap_g = d(y_hat)
+            # for j in range(len(fmap_r)):
+            #     print(i,j,y.shape,y_hat.shape,fmap_r[j].shape,fmap_g[j].shape)
+            y_d_rs.append(y_d_r)
+            y_d_gs.append(y_d_g)
+            fmap_rs.append(fmap_r)
+            fmap_gs.append(fmap_g)
+
+        return y_d_rs, y_d_gs, fmap_rs, fmap_gs
+
+
+class DiscriminatorS(torch.nn.Module):
+    def __init__(self, use_spectral_norm=False):
+        super(DiscriminatorS, self).__init__()
+        norm_f = weight_norm if use_spectral_norm == False else spectral_norm
+        self.convs = nn.ModuleList(
+            [
+                norm_f(Conv1d(1, 16, 15, 1, padding=7)),
+                norm_f(Conv1d(16, 64, 41, 4, groups=4, padding=20)),
+                norm_f(Conv1d(64, 256, 41, 4, groups=16, padding=20)),
+                norm_f(Conv1d(256, 1024, 41, 4, groups=64, padding=20)),
+                norm_f(Conv1d(1024, 1024, 41, 4, groups=256, padding=20)),
+                norm_f(Conv1d(1024, 1024, 5, 1, padding=2)),
+            ]
+        )
+        self.conv_post = norm_f(Conv1d(1024, 1, 3, 1, padding=1))
+
+    def forward(self, x):
+        fmap = []
+
+        for l in self.convs:
+            x = l(x)
+            x = F.leaky_relu(x, modules.LRELU_SLOPE)
+            fmap.append(x)
+        x = self.conv_post(x)
+        fmap.append(x)
+        x = torch.flatten(x, 1, -1)
+
+        return x, fmap
+
+
+class DiscriminatorP(torch.nn.Module):
+    def __init__(self, period, kernel_size=5, stride=3, use_spectral_norm=False):
+        super(DiscriminatorP, self).__init__()
+        self.period = period
+        self.use_spectral_norm = use_spectral_norm
+        norm_f = weight_norm if use_spectral_norm == False else spectral_norm
+        self.convs = nn.ModuleList(
+            [
+                norm_f(
+                    Conv2d(
+                        1,
+                        32,
+                        (kernel_size, 1),
+                        (stride, 1),
+                        padding=(get_padding(kernel_size, 1), 0),
+                    )
+                ),
+                norm_f(
+                    Conv2d(
+                        32,
+                        128,
+                        (kernel_size, 1),
+                        (stride, 1),
+                        padding=(get_padding(kernel_size, 1), 0),
+                    )
+                ),
+                norm_f(
+                    Conv2d(
+                        128,
+                        512,
+                        (kernel_size, 1),
+                        (stride, 1),
+                        padding=(get_padding(kernel_size, 1), 0),
+                    )
+                ),
+                norm_f(
+                    Conv2d(
+                        512,
+                        1024,
+                        (kernel_size, 1),
+                        (stride, 1),
+                        padding=(get_padding(kernel_size, 1), 0),
+                    )
+                ),
+                norm_f(
+                    Conv2d(
+                        1024,
+                        1024,
+                        (kernel_size, 1),
+                        1,
+                        padding=(get_padding(kernel_size, 1), 0),
+                    )
+                ),
+            ]
+        )
+        self.conv_post = norm_f(Conv2d(1024, 1, (3, 1), 1, padding=(1, 0)))
+
+    def forward(self, x):
+        fmap = []
+
+        # 1d to 2d
+        b, c, t = x.shape
+        if t % self.period != 0:  # pad first
+            n_pad = self.period - (t % self.period)
+            x = F.pad(x, (0, n_pad), "reflect")
+            t = t + n_pad
+        x = x.view(b, c, t // self.period, self.period)
+
+        for l in self.convs:
+            x = l(x)
+            x = F.leaky_relu(x, modules.LRELU_SLOPE)
+            fmap.append(x)
+        x = self.conv_post(x)
+        fmap.append(x)
+        x = torch.flatten(x, 1, -1)
+
+        return x, fmap

lib/infer_pack/models_dml.py

@@ -0,0 +1,1124 @@
+import math, pdb, os
+from time import time as ttime
+import torch
+from torch import nn
+from torch.nn import functional as F
+from lib.infer_pack import modules
+from lib.infer_pack import attentions
+from lib.infer_pack import commons
+from lib.infer_pack.commons import init_weights, get_padding
+from torch.nn import Conv1d, ConvTranspose1d, AvgPool1d, Conv2d
+from torch.nn.utils import weight_norm, remove_weight_norm, spectral_norm
+from lib.infer_pack.commons import init_weights
+import numpy as np
+from lib.infer_pack import commons
+
+
+class TextEncoder256(nn.Module):
+    def __init__(
+        self,
+        out_channels,
+        hidden_channels,
+        filter_channels,
+        n_heads,
+        n_layers,
+        kernel_size,
+        p_dropout,
+        f0=True,
+    ):
+        super().__init__()
+        self.out_channels = out_channels
+        self.hidden_channels = hidden_channels
+        self.filter_channels = filter_channels
+        self.n_heads = n_heads
+        self.n_layers = n_layers
+        self.kernel_size = kernel_size
+        self.p_dropout = p_dropout
+        self.emb_phone = nn.Linear(256, hidden_channels)
+        self.lrelu = nn.LeakyReLU(0.1, inplace=True)
+        if f0 == True:
+            self.emb_pitch = nn.Embedding(256, hidden_channels)  # pitch 256
+        self.encoder = attentions.Encoder(
+            hidden_channels, filter_channels, n_heads, n_layers, kernel_size, p_dropout
+        )
+        self.proj = nn.Conv1d(hidden_channels, out_channels * 2, 1)
+
+    def forward(self, phone, pitch, lengths):
+        if pitch == None:
+            x = self.emb_phone(phone)
+        else:
+            x = self.emb_phone(phone) + self.emb_pitch(pitch)
+        x = x * math.sqrt(self.hidden_channels)  # [b, t, h]
+        x = self.lrelu(x)
+        x = torch.transpose(x, 1, -1)  # [b, h, t]
+        x_mask = torch.unsqueeze(commons.sequence_mask(lengths, x.size(2)), 1).to(
+            x.dtype
+        )
+        x = self.encoder(x * x_mask, x_mask)
+        stats = self.proj(x) * x_mask
+
+        m, logs = torch.split(stats, self.out_channels, dim=1)
+        return m, logs, x_mask
+
+
+class TextEncoder768(nn.Module):
+    def __init__(
+        self,
+        out_channels,
+        hidden_channels,
+        filter_channels,
+        n_heads,
+        n_layers,
+        kernel_size,
+        p_dropout,
+        f0=True,
+    ):
+        super().__init__()
+        self.out_channels = out_channels
+        self.hidden_channels = hidden_channels
+        self.filter_channels = filter_channels
+        self.n_heads = n_heads
+        self.n_layers = n_layers
+        self.kernel_size = kernel_size
+        self.p_dropout = p_dropout
+        self.emb_phone = nn.Linear(768, hidden_channels)
+        self.lrelu = nn.LeakyReLU(0.1, inplace=True)
+        if f0 == True:
+            self.emb_pitch = nn.Embedding(256, hidden_channels)  # pitch 256
+        self.encoder = attentions.Encoder(
+            hidden_channels, filter_channels, n_heads, n_layers, kernel_size, p_dropout
+        )
+        self.proj = nn.Conv1d(hidden_channels, out_channels * 2, 1)
+
+    def forward(self, phone, pitch, lengths):
+        if pitch == None:
+            x = self.emb_phone(phone)
+        else:
+            x = self.emb_phone(phone) + self.emb_pitch(pitch)
+        x = x * math.sqrt(self.hidden_channels)  # [b, t, h]
+        x = self.lrelu(x)
+        x = torch.transpose(x, 1, -1)  # [b, h, t]
+        x_mask = torch.unsqueeze(commons.sequence_mask(lengths, x.size(2)), 1).to(
+            x.dtype
+        )
+        x = self.encoder(x * x_mask, x_mask)
+        stats = self.proj(x) * x_mask
+
+        m, logs = torch.split(stats, self.out_channels, dim=1)
+        return m, logs, x_mask
+
+
+class ResidualCouplingBlock(nn.Module):
+    def __init__(
+        self,
+        channels,
+        hidden_channels,
+        kernel_size,
+        dilation_rate,
+        n_layers,
+        n_flows=4,
+        gin_channels=0,
+    ):
+        super().__init__()
+        self.channels = channels
+        self.hidden_channels = hidden_channels
+        self.kernel_size = kernel_size
+        self.dilation_rate = dilation_rate
+        self.n_layers = n_layers
+        self.n_flows = n_flows
+        self.gin_channels = gin_channels
+
+        self.flows = nn.ModuleList()
+        for i in range(n_flows):
+            self.flows.append(
+                modules.ResidualCouplingLayer(
+                    channels,
+                    hidden_channels,
+                    kernel_size,
+                    dilation_rate,
+                    n_layers,
+                    gin_channels=gin_channels,
+                    mean_only=True,
+                )
+            )
+            self.flows.append(modules.Flip())
+
+    def forward(self, x, x_mask, g=None, reverse=False):
+        if not reverse:
+            for flow in self.flows:
+                x, _ = flow(x, x_mask, g=g, reverse=reverse)
+        else:
+            for flow in reversed(self.flows):
+                x = flow(x, x_mask, g=g, reverse=reverse)
+        return x
+
+    def remove_weight_norm(self):
+        for i in range(self.n_flows):
+            self.flows[i * 2].remove_weight_norm()
+
+
+class PosteriorEncoder(nn.Module):
+    def __init__(
+        self,
+        in_channels,
+        out_channels,
+        hidden_channels,
+        kernel_size,
+        dilation_rate,
+        n_layers,
+        gin_channels=0,
+    ):
+        super().__init__()
+        self.in_channels = in_channels
+        self.out_channels = out_channels
+        self.hidden_channels = hidden_channels
+        self.kernel_size = kernel_size
+        self.dilation_rate = dilation_rate
+        self.n_layers = n_layers
+        self.gin_channels = gin_channels
+
+        self.pre = nn.Conv1d(in_channels, hidden_channels, 1)
+        self.enc = modules.WN(
+            hidden_channels,
+            kernel_size,
+            dilation_rate,
+            n_layers,
+            gin_channels=gin_channels,
+        )
+        self.proj = nn.Conv1d(hidden_channels, out_channels * 2, 1)
+
+    def forward(self, x, x_lengths, g=None):
+        x_mask = torch.unsqueeze(commons.sequence_mask(x_lengths, x.size(2)), 1).to(
+            x.dtype
+        )
+        x = self.pre(x) * x_mask
+        x = self.enc(x, x_mask, g=g)
+        stats = self.proj(x) * x_mask
+        m, logs = torch.split(stats, self.out_channels, dim=1)
+        z = (m + torch.randn_like(m) * torch.exp(logs)) * x_mask
+        return z, m, logs, x_mask
+
+    def remove_weight_norm(self):
+        self.enc.remove_weight_norm()
+
+
+class Generator(torch.nn.Module):
+    def __init__(
+        self,
+        initial_channel,
+        resblock,
+        resblock_kernel_sizes,
+        resblock_dilation_sizes,
+        upsample_rates,
+        upsample_initial_channel,
+        upsample_kernel_sizes,
+        gin_channels=0,
+    ):
+        super(Generator, self).__init__()
+        self.num_kernels = len(resblock_kernel_sizes)
+        self.num_upsamples = len(upsample_rates)
+        self.conv_pre = Conv1d(
+            initial_channel, upsample_initial_channel, 7, 1, padding=3
+        )
+        resblock = modules.ResBlock1 if resblock == "1" else modules.ResBlock2
+
+        self.ups = nn.ModuleList()
+        for i, (u, k) in enumerate(zip(upsample_rates, upsample_kernel_sizes)):
+            self.ups.append(
+                weight_norm(
+                    ConvTranspose1d(
+                        upsample_initial_channel // (2**i),
+                        upsample_initial_channel // (2 ** (i + 1)),
+                        k,
+                        u,
+                        padding=(k - u) // 2,
+                    )
+                )
+            )
+
+        self.resblocks = nn.ModuleList()
+        for i in range(len(self.ups)):
+            ch = upsample_initial_channel // (2 ** (i + 1))
+            for j, (k, d) in enumerate(
+                zip(resblock_kernel_sizes, resblock_dilation_sizes)
+            ):
+                self.resblocks.append(resblock(ch, k, d))
+
+        self.conv_post = Conv1d(ch, 1, 7, 1, padding=3, bias=False)
+        self.ups.apply(init_weights)
+
+        if gin_channels != 0:
+            self.cond = nn.Conv1d(gin_channels, upsample_initial_channel, 1)
+
+    def forward(self, x, g=None):
+        x = self.conv_pre(x)
+        if g is not None:
+            x = x + self.cond(g)
+
+        for i in range(self.num_upsamples):
+            x = F.leaky_relu(x, modules.LRELU_SLOPE)
+            x = self.ups[i](x)
+            xs = None
+            for j in range(self.num_kernels):
+                if xs is None:
+                    xs = self.resblocks[i * self.num_kernels + j](x)
+                else:
+                    xs += self.resblocks[i * self.num_kernels + j](x)
+            x = xs / self.num_kernels
+        x = F.leaky_relu(x)
+        x = self.conv_post(x)
+        x = torch.tanh(x)
+
+        return x
+
+    def remove_weight_norm(self):
+        for l in self.ups:
+            remove_weight_norm(l)
+        for l in self.resblocks:
+            l.remove_weight_norm()
+
+
+class SineGen(torch.nn.Module):
+    """Definition of sine generator
+    SineGen(samp_rate, harmonic_num = 0,
+            sine_amp = 0.1, noise_std = 0.003,
+            voiced_threshold = 0,
+            flag_for_pulse=False)
+    samp_rate: sampling rate in Hz
+    harmonic_num: number of harmonic overtones (default 0)
+    sine_amp: amplitude of sine-wavefrom (default 0.1)
+    noise_std: std of Gaussian noise (default 0.003)
+    voiced_thoreshold: F0 threshold for U/V classification (default 0)
+    flag_for_pulse: this SinGen is used inside PulseGen (default False)
+    Note: when flag_for_pulse is True, the first time step of a voiced
+        segment is always sin(np.pi) or cos(0)
+    """
+
+    def __init__(
+        self,
+        samp_rate,
+        harmonic_num=0,
+        sine_amp=0.1,
+        noise_std=0.003,
+        voiced_threshold=0,
+        flag_for_pulse=False,
+    ):
+        super(SineGen, self).__init__()
+        self.sine_amp = sine_amp
+        self.noise_std = noise_std
+        self.harmonic_num = harmonic_num
+        self.dim = self.harmonic_num + 1
+        self.sampling_rate = samp_rate
+        self.voiced_threshold = voiced_threshold
+
+    def _f02uv(self, f0):
+        # generate uv signal
+        uv = torch.ones_like(f0)
+        uv = uv * (f0 > self.voiced_threshold)
+        return uv.float()
+
+    def forward(self, f0, upp):
+        """sine_tensor, uv = forward(f0)
+        input F0: tensor(batchsize=1, length, dim=1)
+                  f0 for unvoiced steps should be 0
+        output sine_tensor: tensor(batchsize=1, length, dim)
+        output uv: tensor(batchsize=1, length, 1)
+        """
+        with torch.no_grad():
+            f0 = f0[:, None].transpose(1, 2)
+            f0_buf = torch.zeros(f0.shape[0], f0.shape[1], self.dim, device=f0.device)
+            # fundamental component
+            f0_buf[:, :, 0] = f0[:, :, 0]
+            for idx in np.arange(self.harmonic_num):
+                f0_buf[:, :, idx + 1] = f0_buf[:, :, 0] * (
+                    idx + 2
+                )  # idx + 2: the (idx+1)-th overtone, (idx+2)-th harmonic
+            rad_values = (f0_buf / self.sampling_rate) % 1  ###%1意味着n_har的乘积无法后处理优化
+            rand_ini = torch.rand(
+                f0_buf.shape[0], f0_buf.shape[2], device=f0_buf.device
+            )
+            rand_ini[:, 0] = 0
+            rad_values[:, 0, :] = rad_values[:, 0, :] + rand_ini
+            tmp_over_one = torch.cumsum(rad_values, 1)  # % 1  #####%1意味着后面的cumsum无法再优化
+            tmp_over_one *= upp
+            tmp_over_one = F.interpolate(
+                tmp_over_one.transpose(2, 1),
+                scale_factor=upp,
+                mode="linear",
+                align_corners=True,
+            ).transpose(2, 1)
+            rad_values = F.interpolate(
+                rad_values.transpose(2, 1), scale_factor=upp, mode="nearest"
+            ).transpose(
+                2, 1
+            )  #######
+            tmp_over_one %= 1
+            tmp_over_one_idx = (tmp_over_one[:, 1:, :] - tmp_over_one[:, :-1, :]) < 0
+            cumsum_shift = torch.zeros_like(rad_values)
+            cumsum_shift[:, 1:, :] = tmp_over_one_idx * -1.0
+            sine_waves = torch.sin(
+                torch.cumsum(rad_values + cumsum_shift, dim=1) * 2 * np.pi
+            )
+            sine_waves = sine_waves * self.sine_amp
+            uv = self._f02uv(f0)
+            uv = F.interpolate(
+                uv.transpose(2, 1), scale_factor=upp, mode="nearest"
+            ).transpose(2, 1)
+            noise_amp = uv * self.noise_std + (1 - uv) * self.sine_amp / 3
+            noise = noise_amp * torch.randn_like(sine_waves)
+            sine_waves = sine_waves * uv + noise
+        return sine_waves, uv, noise
+
+
+class SourceModuleHnNSF(torch.nn.Module):
+    """SourceModule for hn-nsf
+    SourceModule(sampling_rate, harmonic_num=0, sine_amp=0.1,
+                 add_noise_std=0.003, voiced_threshod=0)
+    sampling_rate: sampling_rate in Hz
+    harmonic_num: number of harmonic above F0 (default: 0)
+    sine_amp: amplitude of sine source signal (default: 0.1)
+    add_noise_std: std of additive Gaussian noise (default: 0.003)
+        note that amplitude of noise in unvoiced is decided
+        by sine_amp
+    voiced_threshold: threhold to set U/V given F0 (default: 0)
+    Sine_source, noise_source = SourceModuleHnNSF(F0_sampled)
+    F0_sampled (batchsize, length, 1)
+    Sine_source (batchsize, length, 1)
+    noise_source (batchsize, length 1)
+    uv (batchsize, length, 1)
+    """
+
+    def __init__(
+        self,
+        sampling_rate,
+        harmonic_num=0,
+        sine_amp=0.1,
+        add_noise_std=0.003,
+        voiced_threshod=0,
+        is_half=True,
+    ):
+        super(SourceModuleHnNSF, self).__init__()
+
+        self.sine_amp = sine_amp
+        self.noise_std = add_noise_std
+        self.is_half = is_half
+        # to produce sine waveforms
+        self.l_sin_gen = SineGen(
+            sampling_rate, harmonic_num, sine_amp, add_noise_std, voiced_threshod
+        )
+
+        # to merge source harmonics into a single excitation
+        self.l_linear = torch.nn.Linear(harmonic_num + 1, 1)
+        self.l_tanh = torch.nn.Tanh()
+
+    def forward(self, x, upp=None):
+        sine_wavs, uv, _ = self.l_sin_gen(x, upp)
+        if self.is_half:
+            sine_wavs = sine_wavs.half()
+        sine_merge = self.l_tanh(self.l_linear(sine_wavs))
+        return sine_merge, None, None  # noise, uv
+
+
+class GeneratorNSF(torch.nn.Module):
+    def __init__(
+        self,
+        initial_channel,
+        resblock,
+        resblock_kernel_sizes,
+        resblock_dilation_sizes,
+        upsample_rates,
+        upsample_initial_channel,
+        upsample_kernel_sizes,
+        gin_channels,
+        sr,
+        is_half=False,
+    ):
+        super(GeneratorNSF, self).__init__()
+        self.num_kernels = len(resblock_kernel_sizes)
+        self.num_upsamples = len(upsample_rates)
+
+        self.f0_upsamp = torch.nn.Upsample(scale_factor=np.prod(upsample_rates))
+        self.m_source = SourceModuleHnNSF(
+            sampling_rate=sr, harmonic_num=0, is_half=is_half
+        )
+        self.noise_convs = nn.ModuleList()
+        self.conv_pre = Conv1d(
+            initial_channel, upsample_initial_channel, 7, 1, padding=3
+        )
+        resblock = modules.ResBlock1 if resblock == "1" else modules.ResBlock2
+
+        self.ups = nn.ModuleList()
+        for i, (u, k) in enumerate(zip(upsample_rates, upsample_kernel_sizes)):
+            c_cur = upsample_initial_channel // (2 ** (i + 1))
+            self.ups.append(
+                weight_norm(
+                    ConvTranspose1d(
+                        upsample_initial_channel // (2**i),
+                        upsample_initial_channel // (2 ** (i + 1)),
+                        k,
+                        u,
+                        padding=(k - u) // 2,
+                    )
+                )
+            )
+            if i + 1 < len(upsample_rates):
+                stride_f0 = np.prod(upsample_rates[i + 1 :])
+                self.noise_convs.append(
+                    Conv1d(
+                        1,
+                        c_cur,
+                        kernel_size=stride_f0 * 2,
+                        stride=stride_f0,
+                        padding=stride_f0 // 2,
+                    )
+                )
+            else:
+                self.noise_convs.append(Conv1d(1, c_cur, kernel_size=1))
+
+        self.resblocks = nn.ModuleList()
+        for i in range(len(self.ups)):
+            ch = upsample_initial_channel // (2 ** (i + 1))
+            for j, (k, d) in enumerate(
+                zip(resblock_kernel_sizes, resblock_dilation_sizes)
+            ):
+                self.resblocks.append(resblock(ch, k, d))
+
+        self.conv_post = Conv1d(ch, 1, 7, 1, padding=3, bias=False)
+        self.ups.apply(init_weights)
+
+        if gin_channels != 0:
+            self.cond = nn.Conv1d(gin_channels, upsample_initial_channel, 1)
+
+        self.upp = np.prod(upsample_rates)
+
+    def forward(self, x, f0, g=None):
+        har_source, noi_source, uv = self.m_source(f0, self.upp)
+        har_source = har_source.transpose(1, 2)
+        x = self.conv_pre(x)
+        if g is not None:
+            x = x + self.cond(g)
+
+        for i in range(self.num_upsamples):
+            x = F.leaky_relu(x, modules.LRELU_SLOPE)
+            x = self.ups[i](x)
+            x_source = self.noise_convs[i](har_source)
+            x = x + x_source
+            xs = None
+            for j in range(self.num_kernels):
+                if xs is None:
+                    xs = self.resblocks[i * self.num_kernels + j](x)
+                else:
+                    xs += self.resblocks[i * self.num_kernels + j](x)
+            x = xs / self.num_kernels
+        x = F.leaky_relu(x)
+        x = self.conv_post(x)
+        x = torch.tanh(x)
+        return x
+
+    def remove_weight_norm(self):
+        for l in self.ups:
+            remove_weight_norm(l)
+        for l in self.resblocks:
+            l.remove_weight_norm()
+
+
+sr2sr = {
+    "32k": 32000,
+    "40k": 40000,
+    "48k": 48000,
+}
+
+
+class SynthesizerTrnMs256NSFsid(nn.Module):
+    def __init__(
+        self,
+        spec_channels,
+        segment_size,
+        inter_channels,
+        hidden_channels,
+        filter_channels,
+        n_heads,
+        n_layers,
+        kernel_size,
+        p_dropout,
+        resblock,
+        resblock_kernel_sizes,
+        resblock_dilation_sizes,
+        upsample_rates,
+        upsample_initial_channel,
+        upsample_kernel_sizes,
+        spk_embed_dim,
+        gin_channels,
+        sr,
+        **kwargs
+    ):
+        super().__init__()
+        if type(sr) == type("strr"):
+            sr = sr2sr[sr]
+        self.spec_channels = spec_channels
+        self.inter_channels = inter_channels
+        self.hidden_channels = hidden_channels
+        self.filter_channels = filter_channels
+        self.n_heads = n_heads
+        self.n_layers = n_layers
+        self.kernel_size = kernel_size
+        self.p_dropout = p_dropout
+        self.resblock = resblock
+        self.resblock_kernel_sizes = resblock_kernel_sizes
+        self.resblock_dilation_sizes = resblock_dilation_sizes
+        self.upsample_rates = upsample_rates
+        self.upsample_initial_channel = upsample_initial_channel
+        self.upsample_kernel_sizes = upsample_kernel_sizes
+        self.segment_size = segment_size
+        self.gin_channels = gin_channels
+        # self.hop_length = hop_length#
+        self.spk_embed_dim = spk_embed_dim
+        self.enc_p = TextEncoder256(
+            inter_channels,
+            hidden_channels,
+            filter_channels,
+            n_heads,
+            n_layers,
+            kernel_size,
+            p_dropout,
+        )
+        self.dec = GeneratorNSF(
+            inter_channels,
+            resblock,
+            resblock_kernel_sizes,
+            resblock_dilation_sizes,
+            upsample_rates,
+            upsample_initial_channel,
+            upsample_kernel_sizes,
+            gin_channels=gin_channels,
+            sr=sr,
+            is_half=kwargs["is_half"],
+        )
+        self.enc_q = PosteriorEncoder(
+            spec_channels,
+            inter_channels,
+            hidden_channels,
+            5,
+            1,
+            16,
+            gin_channels=gin_channels,
+        )
+        self.flow = ResidualCouplingBlock(
+            inter_channels, hidden_channels, 5, 1, 3, gin_channels=gin_channels
+        )
+        self.emb_g = nn.Embedding(self.spk_embed_dim, gin_channels)
+        print("gin_channels:", gin_channels, "self.spk_embed_dim:", self.spk_embed_dim)
+
+    def remove_weight_norm(self):
+        self.dec.remove_weight_norm()
+        self.flow.remove_weight_norm()
+        self.enc_q.remove_weight_norm()
+
+    def forward(
+        self, phone, phone_lengths, pitch, pitchf, y, y_lengths, ds
+    ):  # 这里ds是id，[bs,1]
+        # print(1,pitch.shape)#[bs,t]
+        g = self.emb_g(ds).unsqueeze(-1)  # [b, 256, 1]##1是t，广播的
+        m_p, logs_p, x_mask = self.enc_p(phone, pitch, phone_lengths)
+        z, m_q, logs_q, y_mask = self.enc_q(y, y_lengths, g=g)
+        z_p = self.flow(z, y_mask, g=g)
+        z_slice, ids_slice = commons.rand_slice_segments(
+            z, y_lengths, self.segment_size
+        )
+        # print(-1,pitchf.shape,ids_slice,self.segment_size,self.hop_length,self.segment_size//self.hop_length)
+        pitchf = commons.slice_segments2(pitchf, ids_slice, self.segment_size)
+        # print(-2,pitchf.shape,z_slice.shape)
+        o = self.dec(z_slice, pitchf, g=g)
+        return o, ids_slice, x_mask, y_mask, (z, z_p, m_p, logs_p, m_q, logs_q)
+
+    def infer(self, phone, phone_lengths, pitch, nsff0, sid, max_len=None):
+        g = self.emb_g(sid).unsqueeze(-1)
+        m_p, logs_p, x_mask = self.enc_p(phone, pitch, phone_lengths)
+        z_p = (m_p + torch.exp(logs_p) * torch.randn_like(m_p) * 0.66666) * x_mask
+        z = self.flow(z_p, x_mask, g=g, reverse=True)
+        o = self.dec((z * x_mask)[:, :, :max_len], nsff0, g=g)
+        return o, x_mask, (z, z_p, m_p, logs_p)
+
+
+class SynthesizerTrnMs768NSFsid(nn.Module):
+    def __init__(
+        self,
+        spec_channels,
+        segment_size,
+        inter_channels,
+        hidden_channels,
+        filter_channels,
+        n_heads,
+        n_layers,
+        kernel_size,
+        p_dropout,
+        resblock,
+        resblock_kernel_sizes,
+        resblock_dilation_sizes,
+        upsample_rates,
+        upsample_initial_channel,
+        upsample_kernel_sizes,
+        spk_embed_dim,
+        gin_channels,
+        sr,
+        **kwargs
+    ):
+        super().__init__()
+        if type(sr) == type("strr"):
+            sr = sr2sr[sr]
+        self.spec_channels = spec_channels
+        self.inter_channels = inter_channels
+        self.hidden_channels = hidden_channels
+        self.filter_channels = filter_channels
+        self.n_heads = n_heads
+        self.n_layers = n_layers
+        self.kernel_size = kernel_size
+        self.p_dropout = p_dropout
+        self.resblock = resblock
+        self.resblock_kernel_sizes = resblock_kernel_sizes
+        self.resblock_dilation_sizes = resblock_dilation_sizes
+        self.upsample_rates = upsample_rates
+        self.upsample_initial_channel = upsample_initial_channel
+        self.upsample_kernel_sizes = upsample_kernel_sizes
+        self.segment_size = segment_size
+        self.gin_channels = gin_channels
+        # self.hop_length = hop_length#
+        self.spk_embed_dim = spk_embed_dim
+        self.enc_p = TextEncoder768(
+            inter_channels,
+            hidden_channels,
+            filter_channels,
+            n_heads,
+            n_layers,
+            kernel_size,
+            p_dropout,
+        )
+        self.dec = GeneratorNSF(
+            inter_channels,
+            resblock,
+            resblock_kernel_sizes,
+            resblock_dilation_sizes,
+            upsample_rates,
+            upsample_initial_channel,
+            upsample_kernel_sizes,
+            gin_channels=gin_channels,
+            sr=sr,
+            is_half=kwargs["is_half"],
+        )
+        self.enc_q = PosteriorEncoder(
+            spec_channels,
+            inter_channels,
+            hidden_channels,
+            5,
+            1,
+            16,
+            gin_channels=gin_channels,
+        )
+        self.flow = ResidualCouplingBlock(
+            inter_channels, hidden_channels, 5, 1, 3, gin_channels=gin_channels
+        )
+        self.emb_g = nn.Embedding(self.spk_embed_dim, gin_channels)
+        print("gin_channels:", gin_channels, "self.spk_embed_dim:", self.spk_embed_dim)
+
+    def remove_weight_norm(self):
+        self.dec.remove_weight_norm()
+        self.flow.remove_weight_norm()
+        self.enc_q.remove_weight_norm()
+
+    def forward(
+        self, phone, phone_lengths, pitch, pitchf, y, y_lengths, ds
+    ):  # 这里ds是id，[bs,1]
+        # print(1,pitch.shape)#[bs,t]
+        g = self.emb_g(ds).unsqueeze(-1)  # [b, 256, 1]##1是t，广播的
+        m_p, logs_p, x_mask = self.enc_p(phone, pitch, phone_lengths)
+        z, m_q, logs_q, y_mask = self.enc_q(y, y_lengths, g=g)
+        z_p = self.flow(z, y_mask, g=g)
+        z_slice, ids_slice = commons.rand_slice_segments(
+            z, y_lengths, self.segment_size
+        )
+        # print(-1,pitchf.shape,ids_slice,self.segment_size,self.hop_length,self.segment_size//self.hop_length)
+        pitchf = commons.slice_segments2(pitchf, ids_slice, self.segment_size)
+        # print(-2,pitchf.shape,z_slice.shape)
+        o = self.dec(z_slice, pitchf, g=g)
+        return o, ids_slice, x_mask, y_mask, (z, z_p, m_p, logs_p, m_q, logs_q)
+
+    def infer(self, phone, phone_lengths, pitch, nsff0, sid, max_len=None):
+        g = self.emb_g(sid).unsqueeze(-1)
+        m_p, logs_p, x_mask = self.enc_p(phone, pitch, phone_lengths)
+        z_p = (m_p + torch.exp(logs_p) * torch.randn_like(m_p) * 0.66666) * x_mask
+        z = self.flow(z_p, x_mask, g=g, reverse=True)
+        o = self.dec((z * x_mask)[:, :, :max_len], nsff0, g=g)
+        return o, x_mask, (z, z_p, m_p, logs_p)
+
+
+class SynthesizerTrnMs256NSFsid_nono(nn.Module):
+    def __init__(
+        self,
+        spec_channels,
+        segment_size,
+        inter_channels,
+        hidden_channels,
+        filter_channels,
+        n_heads,
+        n_layers,
+        kernel_size,
+        p_dropout,
+        resblock,
+        resblock_kernel_sizes,
+        resblock_dilation_sizes,
+        upsample_rates,
+        upsample_initial_channel,
+        upsample_kernel_sizes,
+        spk_embed_dim,
+        gin_channels,
+        sr=None,
+        **kwargs
+    ):
+        super().__init__()
+        self.spec_channels = spec_channels
+        self.inter_channels = inter_channels
+        self.hidden_channels = hidden_channels
+        self.filter_channels = filter_channels
+        self.n_heads = n_heads
+        self.n_layers = n_layers
+        self.kernel_size = kernel_size
+        self.p_dropout = p_dropout
+        self.resblock = resblock
+        self.resblock_kernel_sizes = resblock_kernel_sizes
+        self.resblock_dilation_sizes = resblock_dilation_sizes
+        self.upsample_rates = upsample_rates
+        self.upsample_initial_channel = upsample_initial_channel
+        self.upsample_kernel_sizes = upsample_kernel_sizes
+        self.segment_size = segment_size
+        self.gin_channels = gin_channels
+        # self.hop_length = hop_length#
+        self.spk_embed_dim = spk_embed_dim
+        self.enc_p = TextEncoder256(
+            inter_channels,
+            hidden_channels,
+            filter_channels,
+            n_heads,
+            n_layers,
+            kernel_size,
+            p_dropout,
+            f0=False,
+        )
+        self.dec = Generator(
+            inter_channels,
+            resblock,
+            resblock_kernel_sizes,
+            resblock_dilation_sizes,
+            upsample_rates,
+            upsample_initial_channel,
+            upsample_kernel_sizes,
+            gin_channels=gin_channels,
+        )
+        self.enc_q = PosteriorEncoder(
+            spec_channels,
+            inter_channels,
+            hidden_channels,
+            5,
+            1,
+            16,
+            gin_channels=gin_channels,
+        )
+        self.flow = ResidualCouplingBlock(
+            inter_channels, hidden_channels, 5, 1, 3, gin_channels=gin_channels
+        )
+        self.emb_g = nn.Embedding(self.spk_embed_dim, gin_channels)
+        print("gin_channels:", gin_channels, "self.spk_embed_dim:", self.spk_embed_dim)
+
+    def remove_weight_norm(self):
+        self.dec.remove_weight_norm()
+        self.flow.remove_weight_norm()
+        self.enc_q.remove_weight_norm()
+
+    def forward(self, phone, phone_lengths, y, y_lengths, ds):  # 这里ds是id，[bs,1]
+        g = self.emb_g(ds).unsqueeze(-1)  # [b, 256, 1]##1是t，广播的
+        m_p, logs_p, x_mask = self.enc_p(phone, None, phone_lengths)
+        z, m_q, logs_q, y_mask = self.enc_q(y, y_lengths, g=g)
+        z_p = self.flow(z, y_mask, g=g)
+        z_slice, ids_slice = commons.rand_slice_segments(
+            z, y_lengths, self.segment_size
+        )
+        o = self.dec(z_slice, g=g)
+        return o, ids_slice, x_mask, y_mask, (z, z_p, m_p, logs_p, m_q, logs_q)
+
+    def infer(self, phone, phone_lengths, sid, max_len=None):
+        g = self.emb_g(sid).unsqueeze(-1)
+        m_p, logs_p, x_mask = self.enc_p(phone, None, phone_lengths)
+        z_p = (m_p + torch.exp(logs_p) * torch.randn_like(m_p) * 0.66666) * x_mask
+        z = self.flow(z_p, x_mask, g=g, reverse=True)
+        o = self.dec((z * x_mask)[:, :, :max_len], g=g)
+        return o, x_mask, (z, z_p, m_p, logs_p)
+
+
+class SynthesizerTrnMs768NSFsid_nono(nn.Module):
+    def __init__(
+        self,
+        spec_channels,
+        segment_size,
+        inter_channels,
+        hidden_channels,
+        filter_channels,
+        n_heads,
+        n_layers,
+        kernel_size,
+        p_dropout,
+        resblock,
+        resblock_kernel_sizes,
+        resblock_dilation_sizes,
+        upsample_rates,
+        upsample_initial_channel,
+        upsample_kernel_sizes,
+        spk_embed_dim,
+        gin_channels,
+        sr=None,
+        **kwargs
+    ):
+        super().__init__()
+        self.spec_channels = spec_channels
+        self.inter_channels = inter_channels
+        self.hidden_channels = hidden_channels
+        self.filter_channels = filter_channels
+        self.n_heads = n_heads
+        self.n_layers = n_layers
+        self.kernel_size = kernel_size
+        self.p_dropout = p_dropout
+        self.resblock = resblock
+        self.resblock_kernel_sizes = resblock_kernel_sizes
+        self.resblock_dilation_sizes = resblock_dilation_sizes
+        self.upsample_rates = upsample_rates
+        self.upsample_initial_channel = upsample_initial_channel
+        self.upsample_kernel_sizes = upsample_kernel_sizes
+        self.segment_size = segment_size
+        self.gin_channels = gin_channels
+        # self.hop_length = hop_length#
+        self.spk_embed_dim = spk_embed_dim
+        self.enc_p = TextEncoder768(
+            inter_channels,
+            hidden_channels,
+            filter_channels,
+            n_heads,
+            n_layers,
+            kernel_size,
+            p_dropout,
+            f0=False,
+        )
+        self.dec = Generator(
+            inter_channels,
+            resblock,
+            resblock_kernel_sizes,
+            resblock_dilation_sizes,
+            upsample_rates,
+            upsample_initial_channel,
+            upsample_kernel_sizes,
+            gin_channels=gin_channels,
+        )
+        self.enc_q = PosteriorEncoder(
+            spec_channels,
+            inter_channels,
+            hidden_channels,
+            5,
+            1,
+            16,
+            gin_channels=gin_channels,
+        )
+        self.flow = ResidualCouplingBlock(
+            inter_channels, hidden_channels, 5, 1, 3, gin_channels=gin_channels
+        )
+        self.emb_g = nn.Embedding(self.spk_embed_dim, gin_channels)
+        print("gin_channels:", gin_channels, "self.spk_embed_dim:", self.spk_embed_dim)
+
+    def remove_weight_norm(self):
+        self.dec.remove_weight_norm()
+        self.flow.remove_weight_norm()
+        self.enc_q.remove_weight_norm()
+
+    def forward(self, phone, phone_lengths, y, y_lengths, ds):  # 这里ds是id，[bs,1]
+        g = self.emb_g(ds).unsqueeze(-1)  # [b, 256, 1]##1是t，广播的
+        m_p, logs_p, x_mask = self.enc_p(phone, None, phone_lengths)
+        z, m_q, logs_q, y_mask = self.enc_q(y, y_lengths, g=g)
+        z_p = self.flow(z, y_mask, g=g)
+        z_slice, ids_slice = commons.rand_slice_segments(
+            z, y_lengths, self.segment_size
+        )
+        o = self.dec(z_slice, g=g)
+        return o, ids_slice, x_mask, y_mask, (z, z_p, m_p, logs_p, m_q, logs_q)
+
+    def infer(self, phone, phone_lengths, sid, max_len=None):
+        g = self.emb_g(sid).unsqueeze(-1)
+        m_p, logs_p, x_mask = self.enc_p(phone, None, phone_lengths)
+        z_p = (m_p + torch.exp(logs_p) * torch.randn_like(m_p) * 0.66666) * x_mask
+        z = self.flow(z_p, x_mask, g=g, reverse=True)
+        o = self.dec((z * x_mask)[:, :, :max_len], g=g)
+        return o, x_mask, (z, z_p, m_p, logs_p)
+
+
+class MultiPeriodDiscriminator(torch.nn.Module):
+    def __init__(self, use_spectral_norm=False):
+        super(MultiPeriodDiscriminator, self).__init__()
+        periods = [2, 3, 5, 7, 11, 17]
+        # periods = [3, 5, 7, 11, 17, 23, 37]
+
+        discs = [DiscriminatorS(use_spectral_norm=use_spectral_norm)]
+        discs = discs + [
+            DiscriminatorP(i, use_spectral_norm=use_spectral_norm) for i in periods
+        ]
+        self.discriminators = nn.ModuleList(discs)
+
+    def forward(self, y, y_hat):
+        y_d_rs = []  #
+        y_d_gs = []
+        fmap_rs = []
+        fmap_gs = []
+        for i, d in enumerate(self.discriminators):
+            y_d_r, fmap_r = d(y)
+            y_d_g, fmap_g = d(y_hat)
+            # for j in range(len(fmap_r)):
+            #     print(i,j,y.shape,y_hat.shape,fmap_r[j].shape,fmap_g[j].shape)
+            y_d_rs.append(y_d_r)
+            y_d_gs.append(y_d_g)
+            fmap_rs.append(fmap_r)
+            fmap_gs.append(fmap_g)
+
+        return y_d_rs, y_d_gs, fmap_rs, fmap_gs
+
+
+class MultiPeriodDiscriminatorV2(torch.nn.Module):
+    def __init__(self, use_spectral_norm=False):
+        super(MultiPeriodDiscriminatorV2, self).__init__()
+        # periods = [2, 3, 5, 7, 11, 17]
+        periods = [2, 3, 5, 7, 11, 17, 23, 37]
+
+        discs = [DiscriminatorS(use_spectral_norm=use_spectral_norm)]
+        discs = discs + [
+            DiscriminatorP(i, use_spectral_norm=use_spectral_norm) for i in periods
+        ]
+        self.discriminators = nn.ModuleList(discs)
+
+    def forward(self, y, y_hat):
+        y_d_rs = []  #
+        y_d_gs = []
+        fmap_rs = []
+        fmap_gs = []
+        for i, d in enumerate(self.discriminators):
+            y_d_r, fmap_r = d(y)
+            y_d_g, fmap_g = d(y_hat)
+            # for j in range(len(fmap_r)):
+            #     print(i,j,y.shape,y_hat.shape,fmap_r[j].shape,fmap_g[j].shape)
+            y_d_rs.append(y_d_r)
+            y_d_gs.append(y_d_g)
+            fmap_rs.append(fmap_r)
+            fmap_gs.append(fmap_g)
+
+        return y_d_rs, y_d_gs, fmap_rs, fmap_gs
+
+
+class DiscriminatorS(torch.nn.Module):
+    def __init__(self, use_spectral_norm=False):
+        super(DiscriminatorS, self).__init__()
+        norm_f = weight_norm if use_spectral_norm == False else spectral_norm
+        self.convs = nn.ModuleList(
+            [
+                norm_f(Conv1d(1, 16, 15, 1, padding=7)),
+                norm_f(Conv1d(16, 64, 41, 4, groups=4, padding=20)),
+                norm_f(Conv1d(64, 256, 41, 4, groups=16, padding=20)),
+                norm_f(Conv1d(256, 1024, 41, 4, groups=64, padding=20)),
+                norm_f(Conv1d(1024, 1024, 41, 4, groups=256, padding=20)),
+                norm_f(Conv1d(1024, 1024, 5, 1, padding=2)),
+            ]
+        )
+        self.conv_post = norm_f(Conv1d(1024, 1, 3, 1, padding=1))
+
+    def forward(self, x):
+        fmap = []
+
+        for l in self.convs:
+            x = l(x)
+            x = F.leaky_relu(x, modules.LRELU_SLOPE)
+            fmap.append(x)
+        x = self.conv_post(x)
+        fmap.append(x)
+        x = torch.flatten(x, 1, -1)
+
+        return x, fmap
+
+
+class DiscriminatorP(torch.nn.Module):
+    def __init__(self, period, kernel_size=5, stride=3, use_spectral_norm=False):
+        super(DiscriminatorP, self).__init__()
+        self.period = period
+        self.use_spectral_norm = use_spectral_norm
+        norm_f = weight_norm if use_spectral_norm == False else spectral_norm
+        self.convs = nn.ModuleList(
+            [
+                norm_f(
+                    Conv2d(
+                        1,
+                        32,
+                        (kernel_size, 1),
+                        (stride, 1),
+                        padding=(get_padding(kernel_size, 1), 0),
+                    )
+                ),
+                norm_f(
+                    Conv2d(
+                        32,
+                        128,
+                        (kernel_size, 1),
+                        (stride, 1),
+                        padding=(get_padding(kernel_size, 1), 0),
+                    )
+                ),
+                norm_f(
+                    Conv2d(
+                        128,
+                        512,
+                        (kernel_size, 1),
+                        (stride, 1),
+                        padding=(get_padding(kernel_size, 1), 0),
+                    )
+                ),
+                norm_f(
+                    Conv2d(
+                        512,
+                        1024,
+                        (kernel_size, 1),
+                        (stride, 1),
+                        padding=(get_padding(kernel_size, 1), 0),
+                    )
+                ),
+                norm_f(
+                    Conv2d(
+                        1024,
+                        1024,
+                        (kernel_size, 1),
+                        1,
+                        padding=(get_padding(kernel_size, 1), 0),
+                    )
+                ),
+            ]
+        )
+        self.conv_post = norm_f(Conv2d(1024, 1, (3, 1), 1, padding=(1, 0)))
+
+    def forward(self, x):
+        fmap = []
+
+        # 1d to 2d
+        b, c, t = x.shape
+        if t % self.period != 0:  # pad first
+            n_pad = self.period - (t % self.period)
+            x = F.pad(x, (0, n_pad), "reflect")
+            t = t + n_pad
+        x = x.view(b, c, t // self.period, self.period)
+
+        for l in self.convs:
+            x = l(x)
+            x = F.leaky_relu(x, modules.LRELU_SLOPE)
+            fmap.append(x)
+        x = self.conv_post(x)
+        fmap.append(x)
+        x = torch.flatten(x, 1, -1)
+
+        return x, fmap

lib/infer_pack/models_onnx.py

@@ -0,0 +1,819 @@
+import math, pdb, os
+from time import time as ttime
+import torch
+from torch import nn
+from torch.nn import functional as F
+from lib.infer_pack import modules
+from lib.infer_pack import attentions
+from lib.infer_pack import commons
+from lib.infer_pack.commons import init_weights, get_padding
+from torch.nn import Conv1d, ConvTranspose1d, AvgPool1d, Conv2d
+from torch.nn.utils import weight_norm, remove_weight_norm, spectral_norm
+from lib.infer_pack.commons import init_weights
+import numpy as np
+from lib.infer_pack import commons
+
+
+class TextEncoder256(nn.Module):
+    def __init__(
+        self,
+        out_channels,
+        hidden_channels,
+        filter_channels,
+        n_heads,
+        n_layers,
+        kernel_size,
+        p_dropout,
+        f0=True,
+    ):
+        super().__init__()
+        self.out_channels = out_channels
+        self.hidden_channels = hidden_channels
+        self.filter_channels = filter_channels
+        self.n_heads = n_heads
+        self.n_layers = n_layers
+        self.kernel_size = kernel_size
+        self.p_dropout = p_dropout
+        self.emb_phone = nn.Linear(256, hidden_channels)
+        self.lrelu = nn.LeakyReLU(0.1, inplace=True)
+        if f0 == True:
+            self.emb_pitch = nn.Embedding(256, hidden_channels)  # pitch 256
+        self.encoder = attentions.Encoder(
+            hidden_channels, filter_channels, n_heads, n_layers, kernel_size, p_dropout
+        )
+        self.proj = nn.Conv1d(hidden_channels, out_channels * 2, 1)
+
+    def forward(self, phone, pitch, lengths):
+        if pitch == None:
+            x = self.emb_phone(phone)
+        else:
+            x = self.emb_phone(phone) + self.emb_pitch(pitch)
+        x = x * math.sqrt(self.hidden_channels)  # [b, t, h]
+        x = self.lrelu(x)
+        x = torch.transpose(x, 1, -1)  # [b, h, t]
+        x_mask = torch.unsqueeze(commons.sequence_mask(lengths, x.size(2)), 1).to(
+            x.dtype
+        )
+        x = self.encoder(x * x_mask, x_mask)
+        stats = self.proj(x) * x_mask
+
+        m, logs = torch.split(stats, self.out_channels, dim=1)
+        return m, logs, x_mask
+
+
+class TextEncoder768(nn.Module):
+    def __init__(
+        self,
+        out_channels,
+        hidden_channels,
+        filter_channels,
+        n_heads,
+        n_layers,
+        kernel_size,
+        p_dropout,
+        f0=True,
+    ):
+        super().__init__()
+        self.out_channels = out_channels
+        self.hidden_channels = hidden_channels
+        self.filter_channels = filter_channels
+        self.n_heads = n_heads
+        self.n_layers = n_layers
+        self.kernel_size = kernel_size
+        self.p_dropout = p_dropout
+        self.emb_phone = nn.Linear(768, hidden_channels)
+        self.lrelu = nn.LeakyReLU(0.1, inplace=True)
+        if f0 == True:
+            self.emb_pitch = nn.Embedding(256, hidden_channels)  # pitch 256
+        self.encoder = attentions.Encoder(
+            hidden_channels, filter_channels, n_heads, n_layers, kernel_size, p_dropout
+        )
+        self.proj = nn.Conv1d(hidden_channels, out_channels * 2, 1)
+
+    def forward(self, phone, pitch, lengths):
+        if pitch == None:
+            x = self.emb_phone(phone)
+        else:
+            x = self.emb_phone(phone) + self.emb_pitch(pitch)
+        x = x * math.sqrt(self.hidden_channels)  # [b, t, h]
+        x = self.lrelu(x)
+        x = torch.transpose(x, 1, -1)  # [b, h, t]
+        x_mask = torch.unsqueeze(commons.sequence_mask(lengths, x.size(2)), 1).to(
+            x.dtype
+        )
+        x = self.encoder(x * x_mask, x_mask)
+        stats = self.proj(x) * x_mask
+
+        m, logs = torch.split(stats, self.out_channels, dim=1)
+        return m, logs, x_mask
+
+
+class ResidualCouplingBlock(nn.Module):
+    def __init__(
+        self,
+        channels,
+        hidden_channels,
+        kernel_size,
+        dilation_rate,
+        n_layers,
+        n_flows=4,
+        gin_channels=0,
+    ):
+        super().__init__()
+        self.channels = channels
+        self.hidden_channels = hidden_channels
+        self.kernel_size = kernel_size
+        self.dilation_rate = dilation_rate
+        self.n_layers = n_layers
+        self.n_flows = n_flows
+        self.gin_channels = gin_channels
+
+        self.flows = nn.ModuleList()
+        for i in range(n_flows):
+            self.flows.append(
+                modules.ResidualCouplingLayer(
+                    channels,
+                    hidden_channels,
+                    kernel_size,
+                    dilation_rate,
+                    n_layers,
+                    gin_channels=gin_channels,
+                    mean_only=True,
+                )
+            )
+            self.flows.append(modules.Flip())
+
+    def forward(self, x, x_mask, g=None, reverse=False):
+        if not reverse:
+            for flow in self.flows:
+                x, _ = flow(x, x_mask, g=g, reverse=reverse)
+        else:
+            for flow in reversed(self.flows):
+                x = flow(x, x_mask, g=g, reverse=reverse)
+        return x
+
+    def remove_weight_norm(self):
+        for i in range(self.n_flows):
+            self.flows[i * 2].remove_weight_norm()
+
+
+class PosteriorEncoder(nn.Module):
+    def __init__(
+        self,
+        in_channels,
+        out_channels,
+        hidden_channels,
+        kernel_size,
+        dilation_rate,
+        n_layers,
+        gin_channels=0,
+    ):
+        super().__init__()
+        self.in_channels = in_channels
+        self.out_channels = out_channels
+        self.hidden_channels = hidden_channels
+        self.kernel_size = kernel_size
+        self.dilation_rate = dilation_rate
+        self.n_layers = n_layers
+        self.gin_channels = gin_channels
+
+        self.pre = nn.Conv1d(in_channels, hidden_channels, 1)
+        self.enc = modules.WN(
+            hidden_channels,
+            kernel_size,
+            dilation_rate,
+            n_layers,
+            gin_channels=gin_channels,
+        )
+        self.proj = nn.Conv1d(hidden_channels, out_channels * 2, 1)
+
+    def forward(self, x, x_lengths, g=None):
+        x_mask = torch.unsqueeze(commons.sequence_mask(x_lengths, x.size(2)), 1).to(
+            x.dtype
+        )
+        x = self.pre(x) * x_mask
+        x = self.enc(x, x_mask, g=g)
+        stats = self.proj(x) * x_mask
+        m, logs = torch.split(stats, self.out_channels, dim=1)
+        z = (m + torch.randn_like(m) * torch.exp(logs)) * x_mask
+        return z, m, logs, x_mask
+
+    def remove_weight_norm(self):
+        self.enc.remove_weight_norm()
+
+
+class Generator(torch.nn.Module):
+    def __init__(
+        self,
+        initial_channel,
+        resblock,
+        resblock_kernel_sizes,
+        resblock_dilation_sizes,
+        upsample_rates,
+        upsample_initial_channel,
+        upsample_kernel_sizes,
+        gin_channels=0,
+    ):
+        super(Generator, self).__init__()
+        self.num_kernels = len(resblock_kernel_sizes)
+        self.num_upsamples = len(upsample_rates)
+        self.conv_pre = Conv1d(
+            initial_channel, upsample_initial_channel, 7, 1, padding=3
+        )
+        resblock = modules.ResBlock1 if resblock == "1" else modules.ResBlock2
+
+        self.ups = nn.ModuleList()
+        for i, (u, k) in enumerate(zip(upsample_rates, upsample_kernel_sizes)):
+            self.ups.append(
+                weight_norm(
+                    ConvTranspose1d(
+                        upsample_initial_channel // (2**i),
+                        upsample_initial_channel // (2 ** (i + 1)),
+                        k,
+                        u,
+                        padding=(k - u) // 2,
+                    )
+                )
+            )
+
+        self.resblocks = nn.ModuleList()
+        for i in range(len(self.ups)):
+            ch = upsample_initial_channel // (2 ** (i + 1))
+            for j, (k, d) in enumerate(
+                zip(resblock_kernel_sizes, resblock_dilation_sizes)
+            ):
+                self.resblocks.append(resblock(ch, k, d))
+
+        self.conv_post = Conv1d(ch, 1, 7, 1, padding=3, bias=False)
+        self.ups.apply(init_weights)
+
+        if gin_channels != 0:
+            self.cond = nn.Conv1d(gin_channels, upsample_initial_channel, 1)
+
+    def forward(self, x, g=None):
+        x = self.conv_pre(x)
+        if g is not None:
+            x = x + self.cond(g)
+
+        for i in range(self.num_upsamples):
+            x = F.leaky_relu(x, modules.LRELU_SLOPE)
+            x = self.ups[i](x)
+            xs = None
+            for j in range(self.num_kernels):
+                if xs is None:
+                    xs = self.resblocks[i * self.num_kernels + j](x)
+                else:
+                    xs += self.resblocks[i * self.num_kernels + j](x)
+            x = xs / self.num_kernels
+        x = F.leaky_relu(x)
+        x = self.conv_post(x)
+        x = torch.tanh(x)
+
+        return x
+
+    def remove_weight_norm(self):
+        for l in self.ups:
+            remove_weight_norm(l)
+        for l in self.resblocks:
+            l.remove_weight_norm()
+
+
+class SineGen(torch.nn.Module):
+    """Definition of sine generator
+    SineGen(samp_rate, harmonic_num = 0,
+            sine_amp = 0.1, noise_std = 0.003,
+            voiced_threshold = 0,
+            flag_for_pulse=False)
+    samp_rate: sampling rate in Hz
+    harmonic_num: number of harmonic overtones (default 0)
+    sine_amp: amplitude of sine-wavefrom (default 0.1)
+    noise_std: std of Gaussian noise (default 0.003)
+    voiced_thoreshold: F0 threshold for U/V classification (default 0)
+    flag_for_pulse: this SinGen is used inside PulseGen (default False)
+    Note: when flag_for_pulse is True, the first time step of a voiced
+        segment is always sin(np.pi) or cos(0)
+    """
+
+    def __init__(
+        self,
+        samp_rate,
+        harmonic_num=0,
+        sine_amp=0.1,
+        noise_std=0.003,
+        voiced_threshold=0,
+        flag_for_pulse=False,
+    ):
+        super(SineGen, self).__init__()
+        self.sine_amp = sine_amp
+        self.noise_std = noise_std
+        self.harmonic_num = harmonic_num
+        self.dim = self.harmonic_num + 1
+        self.sampling_rate = samp_rate
+        self.voiced_threshold = voiced_threshold
+
+    def _f02uv(self, f0):
+        # generate uv signal
+        uv = torch.ones_like(f0)
+        uv = uv * (f0 > self.voiced_threshold)
+        return uv
+
+    def forward(self, f0, upp):
+        """sine_tensor, uv = forward(f0)
+        input F0: tensor(batchsize=1, length, dim=1)
+                  f0 for unvoiced steps should be 0
+        output sine_tensor: tensor(batchsize=1, length, dim)
+        output uv: tensor(batchsize=1, length, 1)
+        """
+        with torch.no_grad():
+            f0 = f0[:, None].transpose(1, 2)
+            f0_buf = torch.zeros(f0.shape[0], f0.shape[1], self.dim, device=f0.device)
+            # fundamental component
+            f0_buf[:, :, 0] = f0[:, :, 0]
+            for idx in np.arange(self.harmonic_num):
+                f0_buf[:, :, idx + 1] = f0_buf[:, :, 0] * (
+                    idx + 2
+                )  # idx + 2: the (idx+1)-th overtone, (idx+2)-th harmonic
+            rad_values = (f0_buf / self.sampling_rate) % 1  ###%1意味着n_har的乘积无法后处理优化
+            rand_ini = torch.rand(
+                f0_buf.shape[0], f0_buf.shape[2], device=f0_buf.device
+            )
+            rand_ini[:, 0] = 0
+            rad_values[:, 0, :] = rad_values[:, 0, :] + rand_ini
+            tmp_over_one = torch.cumsum(rad_values, 1)  # % 1  #####%1意味着后面的cumsum无法再优化
+            tmp_over_one *= upp
+            tmp_over_one = F.interpolate(
+                tmp_over_one.transpose(2, 1),
+                scale_factor=upp,
+                mode="linear",
+                align_corners=True,
+            ).transpose(2, 1)
+            rad_values = F.interpolate(
+                rad_values.transpose(2, 1), scale_factor=upp, mode="nearest"
+            ).transpose(
+                2, 1
+            )  #######
+            tmp_over_one %= 1
+            tmp_over_one_idx = (tmp_over_one[:, 1:, :] - tmp_over_one[:, :-1, :]) < 0
+            cumsum_shift = torch.zeros_like(rad_values)
+            cumsum_shift[:, 1:, :] = tmp_over_one_idx * -1.0
+            sine_waves = torch.sin(
+                torch.cumsum(rad_values + cumsum_shift, dim=1) * 2 * np.pi
+            )
+            sine_waves = sine_waves * self.sine_amp
+            uv = self._f02uv(f0)
+            uv = F.interpolate(
+                uv.transpose(2, 1), scale_factor=upp, mode="nearest"
+            ).transpose(2, 1)
+            noise_amp = uv * self.noise_std + (1 - uv) * self.sine_amp / 3
+            noise = noise_amp * torch.randn_like(sine_waves)
+            sine_waves = sine_waves * uv + noise
+        return sine_waves, uv, noise
+
+
+class SourceModuleHnNSF(torch.nn.Module):
+    """SourceModule for hn-nsf
+    SourceModule(sampling_rate, harmonic_num=0, sine_amp=0.1,
+                 add_noise_std=0.003, voiced_threshod=0)
+    sampling_rate: sampling_rate in Hz
+    harmonic_num: number of harmonic above F0 (default: 0)
+    sine_amp: amplitude of sine source signal (default: 0.1)
+    add_noise_std: std of additive Gaussian noise (default: 0.003)
+        note that amplitude of noise in unvoiced is decided
+        by sine_amp
+    voiced_threshold: threhold to set U/V given F0 (default: 0)
+    Sine_source, noise_source = SourceModuleHnNSF(F0_sampled)
+    F0_sampled (batchsize, length, 1)
+    Sine_source (batchsize, length, 1)
+    noise_source (batchsize, length 1)
+    uv (batchsize, length, 1)
+    """
+
+    def __init__(
+        self,
+        sampling_rate,
+        harmonic_num=0,
+        sine_amp=0.1,
+        add_noise_std=0.003,
+        voiced_threshod=0,
+        is_half=True,
+    ):
+        super(SourceModuleHnNSF, self).__init__()
+
+        self.sine_amp = sine_amp
+        self.noise_std = add_noise_std
+        self.is_half = is_half
+        # to produce sine waveforms
+        self.l_sin_gen = SineGen(
+            sampling_rate, harmonic_num, sine_amp, add_noise_std, voiced_threshod
+        )
+
+        # to merge source harmonics into a single excitation
+        self.l_linear = torch.nn.Linear(harmonic_num + 1, 1)
+        self.l_tanh = torch.nn.Tanh()
+
+    def forward(self, x, upp=None):
+        sine_wavs, uv, _ = self.l_sin_gen(x, upp)
+        if self.is_half:
+            sine_wavs = sine_wavs.half()
+        sine_merge = self.l_tanh(self.l_linear(sine_wavs))
+        return sine_merge, None, None  # noise, uv
+
+
+class GeneratorNSF(torch.nn.Module):
+    def __init__(
+        self,
+        initial_channel,
+        resblock,
+        resblock_kernel_sizes,
+        resblock_dilation_sizes,
+        upsample_rates,
+        upsample_initial_channel,
+        upsample_kernel_sizes,
+        gin_channels,
+        sr,
+        is_half=False,
+    ):
+        super(GeneratorNSF, self).__init__()
+        self.num_kernels = len(resblock_kernel_sizes)
+        self.num_upsamples = len(upsample_rates)
+
+        self.f0_upsamp = torch.nn.Upsample(scale_factor=np.prod(upsample_rates))
+        self.m_source = SourceModuleHnNSF(
+            sampling_rate=sr, harmonic_num=0, is_half=is_half
+        )
+        self.noise_convs = nn.ModuleList()
+        self.conv_pre = Conv1d(
+            initial_channel, upsample_initial_channel, 7, 1, padding=3
+        )
+        resblock = modules.ResBlock1 if resblock == "1" else modules.ResBlock2
+
+        self.ups = nn.ModuleList()
+        for i, (u, k) in enumerate(zip(upsample_rates, upsample_kernel_sizes)):
+            c_cur = upsample_initial_channel // (2 ** (i + 1))
+            self.ups.append(
+                weight_norm(
+                    ConvTranspose1d(
+                        upsample_initial_channel // (2**i),
+                        upsample_initial_channel // (2 ** (i + 1)),
+                        k,
+                        u,
+                        padding=(k - u) // 2,
+                    )
+                )
+            )
+            if i + 1 < len(upsample_rates):
+                stride_f0 = np.prod(upsample_rates[i + 1 :])
+                self.noise_convs.append(
+                    Conv1d(
+                        1,
+                        c_cur,
+                        kernel_size=stride_f0 * 2,
+                        stride=stride_f0,
+                        padding=stride_f0 // 2,
+                    )
+                )
+            else:
+                self.noise_convs.append(Conv1d(1, c_cur, kernel_size=1))
+
+        self.resblocks = nn.ModuleList()
+        for i in range(len(self.ups)):
+            ch = upsample_initial_channel // (2 ** (i + 1))
+            for j, (k, d) in enumerate(
+                zip(resblock_kernel_sizes, resblock_dilation_sizes)
+            ):
+                self.resblocks.append(resblock(ch, k, d))
+
+        self.conv_post = Conv1d(ch, 1, 7, 1, padding=3, bias=False)
+        self.ups.apply(init_weights)
+
+        if gin_channels != 0:
+            self.cond = nn.Conv1d(gin_channels, upsample_initial_channel, 1)
+
+        self.upp = np.prod(upsample_rates)
+
+    def forward(self, x, f0, g=None):
+        har_source, noi_source, uv = self.m_source(f0, self.upp)
+        har_source = har_source.transpose(1, 2)
+        x = self.conv_pre(x)
+        if g is not None:
+            x = x + self.cond(g)
+
+        for i in range(self.num_upsamples):
+            x = F.leaky_relu(x, modules.LRELU_SLOPE)
+            x = self.ups[i](x)
+            x_source = self.noise_convs[i](har_source)
+            x = x + x_source
+            xs = None
+            for j in range(self.num_kernels):
+                if xs is None:
+                    xs = self.resblocks[i * self.num_kernels + j](x)
+                else:
+                    xs += self.resblocks[i * self.num_kernels + j](x)
+            x = xs / self.num_kernels
+        x = F.leaky_relu(x)
+        x = self.conv_post(x)
+        x = torch.tanh(x)
+        return x
+
+    def remove_weight_norm(self):
+        for l in self.ups:
+            remove_weight_norm(l)
+        for l in self.resblocks:
+            l.remove_weight_norm()
+
+
+sr2sr = {
+    "32k": 32000,
+    "40k": 40000,
+    "48k": 48000,
+}
+
+
+class SynthesizerTrnMsNSFsidM(nn.Module):
+    def __init__(
+        self,
+        spec_channels,
+        segment_size,
+        inter_channels,
+        hidden_channels,
+        filter_channels,
+        n_heads,
+        n_layers,
+        kernel_size,
+        p_dropout,
+        resblock,
+        resblock_kernel_sizes,
+        resblock_dilation_sizes,
+        upsample_rates,
+        upsample_initial_channel,
+        upsample_kernel_sizes,
+        spk_embed_dim,
+        gin_channels,
+        sr,
+        version,
+        **kwargs
+    ):
+        super().__init__()
+        if type(sr) == type("strr"):
+            sr = sr2sr[sr]
+        self.spec_channels = spec_channels
+        self.inter_channels = inter_channels
+        self.hidden_channels = hidden_channels
+        self.filter_channels = filter_channels
+        self.n_heads = n_heads
+        self.n_layers = n_layers
+        self.kernel_size = kernel_size
+        self.p_dropout = p_dropout
+        self.resblock = resblock
+        self.resblock_kernel_sizes = resblock_kernel_sizes
+        self.resblock_dilation_sizes = resblock_dilation_sizes
+        self.upsample_rates = upsample_rates
+        self.upsample_initial_channel = upsample_initial_channel
+        self.upsample_kernel_sizes = upsample_kernel_sizes
+        self.segment_size = segment_size
+        self.gin_channels = gin_channels
+        # self.hop_length = hop_length#
+        self.spk_embed_dim = spk_embed_dim
+        if version == "v1":
+            self.enc_p = TextEncoder256(
+                inter_channels,
+                hidden_channels,
+                filter_channels,
+                n_heads,
+                n_layers,
+                kernel_size,
+                p_dropout,
+            )
+        else:
+            self.enc_p = TextEncoder768(
+                inter_channels,
+                hidden_channels,
+                filter_channels,
+                n_heads,
+                n_layers,
+                kernel_size,
+                p_dropout,
+            )
+        self.dec = GeneratorNSF(
+            inter_channels,
+            resblock,
+            resblock_kernel_sizes,
+            resblock_dilation_sizes,
+            upsample_rates,
+            upsample_initial_channel,
+            upsample_kernel_sizes,
+            gin_channels=gin_channels,
+            sr=sr,
+            is_half=kwargs["is_half"],
+        )
+        self.enc_q = PosteriorEncoder(
+            spec_channels,
+            inter_channels,
+            hidden_channels,
+            5,
+            1,
+            16,
+            gin_channels=gin_channels,
+        )
+        self.flow = ResidualCouplingBlock(
+            inter_channels, hidden_channels, 5, 1, 3, gin_channels=gin_channels
+        )
+        self.emb_g = nn.Embedding(self.spk_embed_dim, gin_channels)
+        self.speaker_map = None
+        print("gin_channels:", gin_channels, "self.spk_embed_dim:", self.spk_embed_dim)
+
+    def remove_weight_norm(self):
+        self.dec.remove_weight_norm()
+        self.flow.remove_weight_norm()
+        self.enc_q.remove_weight_norm()
+
+    def construct_spkmixmap(self, n_speaker):
+        self.speaker_map = torch.zeros((n_speaker, 1, 1, self.gin_channels))
+        for i in range(n_speaker):
+            self.speaker_map[i] = self.emb_g(torch.LongTensor([[i]]))
+        self.speaker_map = self.speaker_map.unsqueeze(0)
+
+    def forward(self, phone, phone_lengths, pitch, nsff0, g, rnd, max_len=None):
+        if self.speaker_map is not None:  # [N, S]  *  [S, B, 1, H]
+            g = g.reshape((g.shape[0], g.shape[1], 1, 1, 1))  # [N, S, B, 1, 1]
+            g = g * self.speaker_map  # [N, S, B, 1, H]
+            g = torch.sum(g, dim=1)  # [N, 1, B, 1, H]
+            g = g.transpose(0, -1).transpose(0, -2).squeeze(0)  # [B, H, N]
+        else:
+            g = g.unsqueeze(0)
+            g = self.emb_g(g).transpose(1, 2)
+
+        m_p, logs_p, x_mask = self.enc_p(phone, pitch, phone_lengths)
+        z_p = (m_p + torch.exp(logs_p) * rnd) * x_mask
+        z = self.flow(z_p, x_mask, g=g, reverse=True)
+        o = self.dec((z * x_mask)[:, :, :max_len], nsff0, g=g)
+        return o
+
+
+class MultiPeriodDiscriminator(torch.nn.Module):
+    def __init__(self, use_spectral_norm=False):
+        super(MultiPeriodDiscriminator, self).__init__()
+        periods = [2, 3, 5, 7, 11, 17]
+        # periods = [3, 5, 7, 11, 17, 23, 37]
+
+        discs = [DiscriminatorS(use_spectral_norm=use_spectral_norm)]
+        discs = discs + [
+            DiscriminatorP(i, use_spectral_norm=use_spectral_norm) for i in periods
+        ]
+        self.discriminators = nn.ModuleList(discs)
+
+    def forward(self, y, y_hat):
+        y_d_rs = []  #
+        y_d_gs = []
+        fmap_rs = []
+        fmap_gs = []
+        for i, d in enumerate(self.discriminators):
+            y_d_r, fmap_r = d(y)
+            y_d_g, fmap_g = d(y_hat)
+            # for j in range(len(fmap_r)):
+            #     print(i,j,y.shape,y_hat.shape,fmap_r[j].shape,fmap_g[j].shape)
+            y_d_rs.append(y_d_r)
+            y_d_gs.append(y_d_g)
+            fmap_rs.append(fmap_r)
+            fmap_gs.append(fmap_g)
+
+        return y_d_rs, y_d_gs, fmap_rs, fmap_gs
+
+
+class MultiPeriodDiscriminatorV2(torch.nn.Module):
+    def __init__(self, use_spectral_norm=False):
+        super(MultiPeriodDiscriminatorV2, self).__init__()
+        # periods = [2, 3, 5, 7, 11, 17]
+        periods = [2, 3, 5, 7, 11, 17, 23, 37]
+
+        discs = [DiscriminatorS(use_spectral_norm=use_spectral_norm)]
+        discs = discs + [
+            DiscriminatorP(i, use_spectral_norm=use_spectral_norm) for i in periods
+        ]
+        self.discriminators = nn.ModuleList(discs)
+
+    def forward(self, y, y_hat):
+        y_d_rs = []  #
+        y_d_gs = []
+        fmap_rs = []
+        fmap_gs = []
+        for i, d in enumerate(self.discriminators):
+            y_d_r, fmap_r = d(y)
+            y_d_g, fmap_g = d(y_hat)
+            # for j in range(len(fmap_r)):
+            #     print(i,j,y.shape,y_hat.shape,fmap_r[j].shape,fmap_g[j].shape)
+            y_d_rs.append(y_d_r)
+            y_d_gs.append(y_d_g)
+            fmap_rs.append(fmap_r)
+            fmap_gs.append(fmap_g)
+
+        return y_d_rs, y_d_gs, fmap_rs, fmap_gs
+
+
+class DiscriminatorS(torch.nn.Module):
+    def __init__(self, use_spectral_norm=False):
+        super(DiscriminatorS, self).__init__()
+        norm_f = weight_norm if use_spectral_norm == False else spectral_norm
+        self.convs = nn.ModuleList(
+            [
+                norm_f(Conv1d(1, 16, 15, 1, padding=7)),
+                norm_f(Conv1d(16, 64, 41, 4, groups=4, padding=20)),
+                norm_f(Conv1d(64, 256, 41, 4, groups=16, padding=20)),
+                norm_f(Conv1d(256, 1024, 41, 4, groups=64, padding=20)),
+                norm_f(Conv1d(1024, 1024, 41, 4, groups=256, padding=20)),
+                norm_f(Conv1d(1024, 1024, 5, 1, padding=2)),
+            ]
+        )
+        self.conv_post = norm_f(Conv1d(1024, 1, 3, 1, padding=1))
+
+    def forward(self, x):
+        fmap = []
+
+        for l in self.convs:
+            x = l(x)
+            x = F.leaky_relu(x, modules.LRELU_SLOPE)
+            fmap.append(x)
+        x = self.conv_post(x)
+        fmap.append(x)
+        x = torch.flatten(x, 1, -1)
+
+        return x, fmap
+
+
+class DiscriminatorP(torch.nn.Module):
+    def __init__(self, period, kernel_size=5, stride=3, use_spectral_norm=False):
+        super(DiscriminatorP, self).__init__()
+        self.period = period
+        self.use_spectral_norm = use_spectral_norm
+        norm_f = weight_norm if use_spectral_norm == False else spectral_norm
+        self.convs = nn.ModuleList(
+            [
+                norm_f(
+                    Conv2d(
+                        1,
+                        32,
+                        (kernel_size, 1),
+                        (stride, 1),
+                        padding=(get_padding(kernel_size, 1), 0),
+                    )
+                ),
+                norm_f(
+                    Conv2d(
+                        32,
+                        128,
+                        (kernel_size, 1),
+                        (stride, 1),
+                        padding=(get_padding(kernel_size, 1), 0),
+                    )
+                ),
+                norm_f(
+                    Conv2d(
+                        128,
+                        512,
+                        (kernel_size, 1),
+                        (stride, 1),
+                        padding=(get_padding(kernel_size, 1), 0),
+                    )
+                ),
+                norm_f(
+                    Conv2d(
+                        512,
+                        1024,
+                        (kernel_size, 1),
+                        (stride, 1),
+                        padding=(get_padding(kernel_size, 1), 0),
+                    )
+                ),
+                norm_f(
+                    Conv2d(
+                        1024,
+                        1024,
+                        (kernel_size, 1),
+                        1,
+                        padding=(get_padding(kernel_size, 1), 0),
+                    )
+                ),
+            ]
+        )
+        self.conv_post = norm_f(Conv2d(1024, 1, (3, 1), 1, padding=(1, 0)))
+
+    def forward(self, x):
+        fmap = []
+
+        # 1d to 2d
+        b, c, t = x.shape
+        if t % self.period != 0:  # pad first
+            n_pad = self.period - (t % self.period)
+            x = F.pad(x, (0, n_pad), "reflect")
+            t = t + n_pad
+        x = x.view(b, c, t // self.period, self.period)
+
+        for l in self.convs:
+            x = l(x)
+            x = F.leaky_relu(x, modules.LRELU_SLOPE)
+            fmap.append(x)
+        x = self.conv_post(x)
+        fmap.append(x)
+        x = torch.flatten(x, 1, -1)
+
+        return x, fmap

lib/infer_pack/modules.py

@@ -0,0 +1,522 @@
+import copy
+import math
+import numpy as np
+import scipy
+import torch
+from torch import nn
+from torch.nn import functional as F
+
+from torch.nn import Conv1d, ConvTranspose1d, AvgPool1d, Conv2d
+from torch.nn.utils import weight_norm, remove_weight_norm
+
+from lib.infer_pack import commons
+from lib.infer_pack.commons import init_weights, get_padding
+from lib.infer_pack.transforms import piecewise_rational_quadratic_transform
+
+
+LRELU_SLOPE = 0.1
+
+
+class LayerNorm(nn.Module):
+    def __init__(self, channels, eps=1e-5):
+        super().__init__()
+        self.channels = channels
+        self.eps = eps
+
+        self.gamma = nn.Parameter(torch.ones(channels))
+        self.beta = nn.Parameter(torch.zeros(channels))
+
+    def forward(self, x):
+        x = x.transpose(1, -1)
+        x = F.layer_norm(x, (self.channels,), self.gamma, self.beta, self.eps)
+        return x.transpose(1, -1)
+
+
+class ConvReluNorm(nn.Module):
+    def __init__(
+        self,
+        in_channels,
+        hidden_channels,
+        out_channels,
+        kernel_size,
+        n_layers,
+        p_dropout,
+    ):
+        super().__init__()
+        self.in_channels = in_channels
+        self.hidden_channels = hidden_channels
+        self.out_channels = out_channels
+        self.kernel_size = kernel_size
+        self.n_layers = n_layers
+        self.p_dropout = p_dropout
+        assert n_layers > 1, "Number of layers should be larger than 0."
+
+        self.conv_layers = nn.ModuleList()
+        self.norm_layers = nn.ModuleList()
+        self.conv_layers.append(
+            nn.Conv1d(
+                in_channels, hidden_channels, kernel_size, padding=kernel_size // 2
+            )
+        )
+        self.norm_layers.append(LayerNorm(hidden_channels))
+        self.relu_drop = nn.Sequential(nn.ReLU(), nn.Dropout(p_dropout))
+        for _ in range(n_layers - 1):
+            self.conv_layers.append(
+                nn.Conv1d(
+                    hidden_channels,
+                    hidden_channels,
+                    kernel_size,
+                    padding=kernel_size // 2,
+                )
+            )
+            self.norm_layers.append(LayerNorm(hidden_channels))
+        self.proj = nn.Conv1d(hidden_channels, out_channels, 1)
+        self.proj.weight.data.zero_()
+        self.proj.bias.data.zero_()
+
+    def forward(self, x, x_mask):
+        x_org = x
+        for i in range(self.n_layers):
+            x = self.conv_layers[i](x * x_mask)
+            x = self.norm_layers[i](x)
+            x = self.relu_drop(x)
+        x = x_org + self.proj(x)
+        return x * x_mask
+
+
+class DDSConv(nn.Module):
+    """
+    Dialted and Depth-Separable Convolution
+    """
+
+    def __init__(self, channels, kernel_size, n_layers, p_dropout=0.0):
+        super().__init__()
+        self.channels = channels
+        self.kernel_size = kernel_size
+        self.n_layers = n_layers
+        self.p_dropout = p_dropout
+
+        self.drop = nn.Dropout(p_dropout)
+        self.convs_sep = nn.ModuleList()
+        self.convs_1x1 = nn.ModuleList()
+        self.norms_1 = nn.ModuleList()
+        self.norms_2 = nn.ModuleList()
+        for i in range(n_layers):
+            dilation = kernel_size**i
+            padding = (kernel_size * dilation - dilation) // 2
+            self.convs_sep.append(
+                nn.Conv1d(
+                    channels,
+                    channels,
+                    kernel_size,
+                    groups=channels,
+                    dilation=dilation,
+                    padding=padding,
+                )
+            )
+            self.convs_1x1.append(nn.Conv1d(channels, channels, 1))
+            self.norms_1.append(LayerNorm(channels))
+            self.norms_2.append(LayerNorm(channels))
+
+    def forward(self, x, x_mask, g=None):
+        if g is not None:
+            x = x + g
+        for i in range(self.n_layers):
+            y = self.convs_sep[i](x * x_mask)
+            y = self.norms_1[i](y)
+            y = F.gelu(y)
+            y = self.convs_1x1[i](y)
+            y = self.norms_2[i](y)
+            y = F.gelu(y)
+            y = self.drop(y)
+            x = x + y
+        return x * x_mask
+
+
+class WN(torch.nn.Module):
+    def __init__(
+        self,
+        hidden_channels,
+        kernel_size,
+        dilation_rate,
+        n_layers,
+        gin_channels=0,
+        p_dropout=0,
+    ):
+        super(WN, self).__init__()
+        assert kernel_size % 2 == 1
+        self.hidden_channels = hidden_channels
+        self.kernel_size = (kernel_size,)
+        self.dilation_rate = dilation_rate
+        self.n_layers = n_layers
+        self.gin_channels = gin_channels
+        self.p_dropout = p_dropout
+
+        self.in_layers = torch.nn.ModuleList()
+        self.res_skip_layers = torch.nn.ModuleList()
+        self.drop = nn.Dropout(p_dropout)
+
+        if gin_channels != 0:
+            cond_layer = torch.nn.Conv1d(
+                gin_channels, 2 * hidden_channels * n_layers, 1
+            )
+            self.cond_layer = torch.nn.utils.weight_norm(cond_layer, name="weight")
+
+        for i in range(n_layers):
+            dilation = dilation_rate**i
+            padding = int((kernel_size * dilation - dilation) / 2)
+            in_layer = torch.nn.Conv1d(
+                hidden_channels,
+                2 * hidden_channels,
+                kernel_size,
+                dilation=dilation,
+                padding=padding,
+            )
+            in_layer = torch.nn.utils.weight_norm(in_layer, name="weight")
+            self.in_layers.append(in_layer)
+
+            # last one is not necessary
+            if i < n_layers - 1:
+                res_skip_channels = 2 * hidden_channels
+            else:
+                res_skip_channels = hidden_channels
+
+            res_skip_layer = torch.nn.Conv1d(hidden_channels, res_skip_channels, 1)
+            res_skip_layer = torch.nn.utils.weight_norm(res_skip_layer, name="weight")
+            self.res_skip_layers.append(res_skip_layer)
+
+    def forward(self, x, x_mask, g=None, **kwargs):
+        output = torch.zeros_like(x)
+        n_channels_tensor = torch.IntTensor([self.hidden_channels])
+
+        if g is not None:
+            g = self.cond_layer(g)
+
+        for i in range(self.n_layers):
+            x_in = self.in_layers[i](x)
+            if g is not None:
+                cond_offset = i * 2 * self.hidden_channels
+                g_l = g[:, cond_offset : cond_offset + 2 * self.hidden_channels, :]
+            else:
+                g_l = torch.zeros_like(x_in)
+
+            acts = commons.fused_add_tanh_sigmoid_multiply(x_in, g_l, n_channels_tensor)
+            acts = self.drop(acts)
+
+            res_skip_acts = self.res_skip_layers[i](acts)
+            if i < self.n_layers - 1:
+                res_acts = res_skip_acts[:, : self.hidden_channels, :]
+                x = (x + res_acts) * x_mask
+                output = output + res_skip_acts[:, self.hidden_channels :, :]
+            else:
+                output = output + res_skip_acts
+        return output * x_mask
+
+    def remove_weight_norm(self):
+        if self.gin_channels != 0:
+            torch.nn.utils.remove_weight_norm(self.cond_layer)
+        for l in self.in_layers:
+            torch.nn.utils.remove_weight_norm(l)
+        for l in self.res_skip_layers:
+            torch.nn.utils.remove_weight_norm(l)
+
+
+class ResBlock1(torch.nn.Module):
+    def __init__(self, channels, kernel_size=3, dilation=(1, 3, 5)):
+        super(ResBlock1, self).__init__()
+        self.convs1 = nn.ModuleList(
+            [
+                weight_norm(
+                    Conv1d(
+                        channels,
+                        channels,
+                        kernel_size,
+                        1,
+                        dilation=dilation[0],
+                        padding=get_padding(kernel_size, dilation[0]),
+                    )
+                ),
+                weight_norm(
+                    Conv1d(
+                        channels,
+                        channels,
+                        kernel_size,
+                        1,
+                        dilation=dilation[1],
+                        padding=get_padding(kernel_size, dilation[1]),
+                    )
+                ),
+                weight_norm(
+                    Conv1d(
+                        channels,
+                        channels,
+                        kernel_size,
+                        1,
+                        dilation=dilation[2],
+                        padding=get_padding(kernel_size, dilation[2]),
+                    )
+                ),
+            ]
+        )
+        self.convs1.apply(init_weights)
+
+        self.convs2 = nn.ModuleList(
+            [
+                weight_norm(
+                    Conv1d(
+                        channels,
+                        channels,
+                        kernel_size,
+                        1,
+                        dilation=1,
+                        padding=get_padding(kernel_size, 1),
+                    )
+                ),
+                weight_norm(
+                    Conv1d(
+                        channels,
+                        channels,
+                        kernel_size,
+                        1,
+                        dilation=1,
+                        padding=get_padding(kernel_size, 1),
+                    )
+                ),
+                weight_norm(
+                    Conv1d(
+                        channels,
+                        channels,
+                        kernel_size,
+                        1,
+                        dilation=1,
+                        padding=get_padding(kernel_size, 1),
+                    )
+                ),
+            ]
+        )
+        self.convs2.apply(init_weights)
+
+    def forward(self, x, x_mask=None):
+        for c1, c2 in zip(self.convs1, self.convs2):
+            xt = F.leaky_relu(x, LRELU_SLOPE)
+            if x_mask is not None:
+                xt = xt * x_mask
+            xt = c1(xt)
+            xt = F.leaky_relu(xt, LRELU_SLOPE)
+            if x_mask is not None:
+                xt = xt * x_mask
+            xt = c2(xt)
+            x = xt + x
+        if x_mask is not None:
+            x = x * x_mask
+        return x
+
+    def remove_weight_norm(self):
+        for l in self.convs1:
+            remove_weight_norm(l)
+        for l in self.convs2:
+            remove_weight_norm(l)
+
+
+class ResBlock2(torch.nn.Module):
+    def __init__(self, channels, kernel_size=3, dilation=(1, 3)):
+        super(ResBlock2, self).__init__()
+        self.convs = nn.ModuleList(
+            [
+                weight_norm(
+                    Conv1d(
+                        channels,
+                        channels,
+                        kernel_size,
+                        1,
+                        dilation=dilation[0],
+                        padding=get_padding(kernel_size, dilation[0]),
+                    )
+                ),
+                weight_norm(
+                    Conv1d(
+                        channels,
+                        channels,
+                        kernel_size,
+                        1,
+                        dilation=dilation[1],
+                        padding=get_padding(kernel_size, dilation[1]),
+                    )
+                ),
+            ]
+        )
+        self.convs.apply(init_weights)
+
+    def forward(self, x, x_mask=None):
+        for c in self.convs:
+            xt = F.leaky_relu(x, LRELU_SLOPE)
+            if x_mask is not None:
+                xt = xt * x_mask
+            xt = c(xt)
+            x = xt + x
+        if x_mask is not None:
+            x = x * x_mask
+        return x
+
+    def remove_weight_norm(self):
+        for l in self.convs:
+            remove_weight_norm(l)
+
+
+class Log(nn.Module):
+    def forward(self, x, x_mask, reverse=False, **kwargs):
+        if not reverse:
+            y = torch.log(torch.clamp_min(x, 1e-5)) * x_mask
+            logdet = torch.sum(-y, [1, 2])
+            return y, logdet
+        else:
+            x = torch.exp(x) * x_mask
+            return x
+
+
+class Flip(nn.Module):
+    def forward(self, x, *args, reverse=False, **kwargs):
+        x = torch.flip(x, [1])
+        if not reverse:
+            logdet = torch.zeros(x.size(0)).to(dtype=x.dtype, device=x.device)
+            return x, logdet
+        else:
+            return x
+
+
+class ElementwiseAffine(nn.Module):
+    def __init__(self, channels):
+        super().__init__()
+        self.channels = channels
+        self.m = nn.Parameter(torch.zeros(channels, 1))
+        self.logs = nn.Parameter(torch.zeros(channels, 1))
+
+    def forward(self, x, x_mask, reverse=False, **kwargs):
+        if not reverse:
+            y = self.m + torch.exp(self.logs) * x
+            y = y * x_mask
+            logdet = torch.sum(self.logs * x_mask, [1, 2])
+            return y, logdet
+        else:
+            x = (x - self.m) * torch.exp(-self.logs) * x_mask
+            return x
+
+
+class ResidualCouplingLayer(nn.Module):
+    def __init__(
+        self,
+        channels,
+        hidden_channels,
+        kernel_size,
+        dilation_rate,
+        n_layers,
+        p_dropout=0,
+        gin_channels=0,
+        mean_only=False,
+    ):
+        assert channels % 2 == 0, "channels should be divisible by 2"
+        super().__init__()
+        self.channels = channels
+        self.hidden_channels = hidden_channels
+        self.kernel_size = kernel_size
+        self.dilation_rate = dilation_rate
+        self.n_layers = n_layers
+        self.half_channels = channels // 2
+        self.mean_only = mean_only
+
+        self.pre = nn.Conv1d(self.half_channels, hidden_channels, 1)
+        self.enc = WN(
+            hidden_channels,
+            kernel_size,
+            dilation_rate,
+            n_layers,
+            p_dropout=p_dropout,
+            gin_channels=gin_channels,
+        )
+        self.post = nn.Conv1d(hidden_channels, self.half_channels * (2 - mean_only), 1)
+        self.post.weight.data.zero_()
+        self.post.bias.data.zero_()
+
+    def forward(self, x, x_mask, g=None, reverse=False):
+        x0, x1 = torch.split(x, [self.half_channels] * 2, 1)
+        h = self.pre(x0) * x_mask
+        h = self.enc(h, x_mask, g=g)
+        stats = self.post(h) * x_mask
+        if not self.mean_only:
+            m, logs = torch.split(stats, [self.half_channels] * 2, 1)
+        else:
+            m = stats
+            logs = torch.zeros_like(m)
+
+        if not reverse:
+            x1 = m + x1 * torch.exp(logs) * x_mask
+            x = torch.cat([x0, x1], 1)
+            logdet = torch.sum(logs, [1, 2])
+            return x, logdet
+        else:
+            x1 = (x1 - m) * torch.exp(-logs) * x_mask
+            x = torch.cat([x0, x1], 1)
+            return x
+
+    def remove_weight_norm(self):
+        self.enc.remove_weight_norm()
+
+
+class ConvFlow(nn.Module):
+    def __init__(
+        self,
+        in_channels,
+        filter_channels,
+        kernel_size,
+        n_layers,
+        num_bins=10,
+        tail_bound=5.0,
+    ):
+        super().__init__()
+        self.in_channels = in_channels
+        self.filter_channels = filter_channels
+        self.kernel_size = kernel_size
+        self.n_layers = n_layers
+        self.num_bins = num_bins
+        self.tail_bound = tail_bound
+        self.half_channels = in_channels // 2
+
+        self.pre = nn.Conv1d(self.half_channels, filter_channels, 1)
+        self.convs = DDSConv(filter_channels, kernel_size, n_layers, p_dropout=0.0)
+        self.proj = nn.Conv1d(
+            filter_channels, self.half_channels * (num_bins * 3 - 1), 1
+        )
+        self.proj.weight.data.zero_()
+        self.proj.bias.data.zero_()
+
+    def forward(self, x, x_mask, g=None, reverse=False):
+        x0, x1 = torch.split(x, [self.half_channels] * 2, 1)
+        h = self.pre(x0)
+        h = self.convs(h, x_mask, g=g)
+        h = self.proj(h) * x_mask
+
+        b, c, t = x0.shape
+        h = h.reshape(b, c, -1, t).permute(0, 1, 3, 2)  # [b, cx?, t] -> [b, c, t, ?]
+
+        unnormalized_widths = h[..., : self.num_bins] / math.sqrt(self.filter_channels)
+        unnormalized_heights = h[..., self.num_bins : 2 * self.num_bins] / math.sqrt(
+            self.filter_channels
+        )
+        unnormalized_derivatives = h[..., 2 * self.num_bins :]
+
+        x1, logabsdet = piecewise_rational_quadratic_transform(
+            x1,
+            unnormalized_widths,
+            unnormalized_heights,
+            unnormalized_derivatives,
+            inverse=reverse,
+            tails="linear",
+            tail_bound=self.tail_bound,
+        )
+
+        x = torch.cat([x0, x1], 1) * x_mask
+        logdet = torch.sum(logabsdet * x_mask, [1, 2])
+        if not reverse:
+            return x, logdet
+        else:
+            return x

lib/infer_pack/modules/F0Predictor/DioF0Predictor.py

@@ -0,0 +1,90 @@
+from lib.infer_pack.modules.F0Predictor.F0Predictor import F0Predictor
+import pyworld
+import numpy as np
+
+
+class DioF0Predictor(F0Predictor):
+    def __init__(self, hop_length=512, f0_min=50, f0_max=1100, sampling_rate=44100):
+        self.hop_length = hop_length
+        self.f0_min = f0_min
+        self.f0_max = f0_max
+        self.sampling_rate = sampling_rate
+
+    def interpolate_f0(self, f0):
+        """
+        对F0进行插值处理
+        """
+
+        data = np.reshape(f0, (f0.size, 1))
+
+        vuv_vector = np.zeros((data.size, 1), dtype=np.float32)
+        vuv_vector[data > 0.0] = 1.0
+        vuv_vector[data <= 0.0] = 0.0
+
+        ip_data = data
+
+        frame_number = data.size
+        last_value = 0.0
+        for i in range(frame_number):
+            if data[i] <= 0.0:
+                j = i + 1
+                for j in range(i + 1, frame_number):
+                    if data[j] > 0.0:
+                        break
+                if j < frame_number - 1:
+                    if last_value > 0.0:
+                        step = (data[j] - data[i - 1]) / float(j - i)
+                        for k in range(i, j):
+                            ip_data[k] = data[i - 1] + step * (k - i + 1)
+                    else:
+                        for k in range(i, j):
+                            ip_data[k] = data[j]
+                else:
+                    for k in range(i, frame_number):
+                        ip_data[k] = last_value
+            else:
+                ip_data[i] = data[i]  # 这里可能存在一个没有必要的拷贝
+                last_value = data[i]
+
+        return ip_data[:, 0], vuv_vector[:, 0]
+
+    def resize_f0(self, x, target_len):
+        source = np.array(x)
+        source[source < 0.001] = np.nan
+        target = np.interp(
+            np.arange(0, len(source) * target_len, len(source)) / target_len,
+            np.arange(0, len(source)),
+            source,
+        )
+        res = np.nan_to_num(target)
+        return res
+
+    def compute_f0(self, wav, p_len=None):
+        if p_len is None:
+            p_len = wav.shape[0] // self.hop_length
+        f0, t = pyworld.dio(
+            wav.astype(np.double),
+            fs=self.sampling_rate,
+            f0_floor=self.f0_min,
+            f0_ceil=self.f0_max,
+            frame_period=1000 * self.hop_length / self.sampling_rate,
+        )
+        f0 = pyworld.stonemask(wav.astype(np.double), f0, t, self.sampling_rate)
+        for index, pitch in enumerate(f0):
+            f0[index] = round(pitch, 1)
+        return self.interpolate_f0(self.resize_f0(f0, p_len))[0]
+
+    def compute_f0_uv(self, wav, p_len=None):
+        if p_len is None:
+            p_len = wav.shape[0] // self.hop_length
+        f0, t = pyworld.dio(
+            wav.astype(np.double),
+            fs=self.sampling_rate,
+            f0_floor=self.f0_min,
+            f0_ceil=self.f0_max,
+            frame_period=1000 * self.hop_length / self.sampling_rate,
+        )
+        f0 = pyworld.stonemask(wav.astype(np.double), f0, t, self.sampling_rate)
+        for index, pitch in enumerate(f0):
+            f0[index] = round(pitch, 1)
+        return self.interpolate_f0(self.resize_f0(f0, p_len))

lib/infer_pack/modules/F0Predictor/F0Predictor.py

@@ -0,0 +1,16 @@
+class F0Predictor(object):
+    def compute_f0(self, wav, p_len):
+        """
+        input: wav:[signal_length]
+               p_len:int
+        output: f0:[signal_length//hop_length]
+        """
+        pass
+
+    def compute_f0_uv(self, wav, p_len):
+        """
+        input: wav:[signal_length]
+               p_len:int
+        output: f0:[signal_length//hop_length],uv:[signal_length//hop_length]
+        """
+        pass

lib/infer_pack/modules/F0Predictor/HarvestF0Predictor.py

@@ -0,0 +1,86 @@
+from lib.infer_pack.modules.F0Predictor.F0Predictor import F0Predictor
+import pyworld
+import numpy as np
+
+
+class HarvestF0Predictor(F0Predictor):
+    def __init__(self, hop_length=512, f0_min=50, f0_max=1100, sampling_rate=44100):
+        self.hop_length = hop_length
+        self.f0_min = f0_min
+        self.f0_max = f0_max
+        self.sampling_rate = sampling_rate
+
+    def interpolate_f0(self, f0):
+        """
+        对F0进行插值处理
+        """
+
+        data = np.reshape(f0, (f0.size, 1))
+
+        vuv_vector = np.zeros((data.size, 1), dtype=np.float32)
+        vuv_vector[data > 0.0] = 1.0
+        vuv_vector[data <= 0.0] = 0.0
+
+        ip_data = data
+
+        frame_number = data.size
+        last_value = 0.0
+        for i in range(frame_number):
+            if data[i] <= 0.0:
+                j = i + 1
+                for j in range(i + 1, frame_number):
+                    if data[j] > 0.0:
+                        break
+                if j < frame_number - 1:
+                    if last_value > 0.0:
+                        step = (data[j] - data[i - 1]) / float(j - i)
+                        for k in range(i, j):
+                            ip_data[k] = data[i - 1] + step * (k - i + 1)
+                    else:
+                        for k in range(i, j):
+                            ip_data[k] = data[j]
+                else:
+                    for k in range(i, frame_number):
+                        ip_data[k] = last_value
+            else:
+                ip_data[i] = data[i]  # 这里可能存在一个没有必要的拷贝
+                last_value = data[i]
+
+        return ip_data[:, 0], vuv_vector[:, 0]
+
+    def resize_f0(self, x, target_len):
+        source = np.array(x)
+        source[source < 0.001] = np.nan
+        target = np.interp(
+            np.arange(0, len(source) * target_len, len(source)) / target_len,
+            np.arange(0, len(source)),
+            source,
+        )
+        res = np.nan_to_num(target)
+        return res
+
+    def compute_f0(self, wav, p_len=None):
+        if p_len is None:
+            p_len = wav.shape[0] // self.hop_length
+        f0, t = pyworld.harvest(
+            wav.astype(np.double),
+            fs=self.hop_length,
+            f0_ceil=self.f0_max,
+            f0_floor=self.f0_min,
+            frame_period=1000 * self.hop_length / self.sampling_rate,
+        )
+        f0 = pyworld.stonemask(wav.astype(np.double), f0, t, self.fs)
+        return self.interpolate_f0(self.resize_f0(f0, p_len))[0]
+
+    def compute_f0_uv(self, wav, p_len=None):
+        if p_len is None:
+            p_len = wav.shape[0] // self.hop_length
+        f0, t = pyworld.harvest(
+            wav.astype(np.double),
+            fs=self.sampling_rate,
+            f0_floor=self.f0_min,
+            f0_ceil=self.f0_max,
+            frame_period=1000 * self.hop_length / self.sampling_rate,
+        )
+        f0 = pyworld.stonemask(wav.astype(np.double), f0, t, self.sampling_rate)
+        return self.interpolate_f0(self.resize_f0(f0, p_len))

lib/infer_pack/modules/F0Predictor/PMF0Predictor.py

@@ -0,0 +1,97 @@
+from lib.infer_pack.modules.F0Predictor.F0Predictor import F0Predictor
+import parselmouth
+import numpy as np
+
+
+class PMF0Predictor(F0Predictor):
+    def __init__(self, hop_length=512, f0_min=50, f0_max=1100, sampling_rate=44100):
+        self.hop_length = hop_length
+        self.f0_min = f0_min
+        self.f0_max = f0_max
+        self.sampling_rate = sampling_rate
+
+    def interpolate_f0(self, f0):
+        """
+        对F0进行插值处理
+        """
+
+        data = np.reshape(f0, (f0.size, 1))
+
+        vuv_vector = np.zeros((data.size, 1), dtype=np.float32)
+        vuv_vector[data > 0.0] = 1.0
+        vuv_vector[data <= 0.0] = 0.0
+
+        ip_data = data
+
+        frame_number = data.size
+        last_value = 0.0
+        for i in range(frame_number):
+            if data[i] <= 0.0:
+                j = i + 1
+                for j in range(i + 1, frame_number):
+                    if data[j] > 0.0:
+                        break
+                if j < frame_number - 1:
+                    if last_value > 0.0:
+                        step = (data[j] - data[i - 1]) / float(j - i)
+                        for k in range(i, j):
+                            ip_data[k] = data[i - 1] + step * (k - i + 1)
+                    else:
+                        for k in range(i, j):
+                            ip_data[k] = data[j]
+                else:
+                    for k in range(i, frame_number):
+                        ip_data[k] = last_value
+            else:
+                ip_data[i] = data[i]  # 这里可能存在一个没有必要的拷贝
+                last_value = data[i]
+
+        return ip_data[:, 0], vuv_vector[:, 0]
+
+    def compute_f0(self, wav, p_len=None):
+        x = wav
+        if p_len is None:
+            p_len = x.shape[0] // self.hop_length
+        else:
+            assert abs(p_len - x.shape[0] // self.hop_length) < 4, "pad length error"
+        time_step = self.hop_length / self.sampling_rate * 1000
+        f0 = (
+            parselmouth.Sound(x, self.sampling_rate)
+            .to_pitch_ac(
+                time_step=time_step / 1000,
+                voicing_threshold=0.6,
+                pitch_floor=self.f0_min,
+                pitch_ceiling=self.f0_max,
+            )
+            .selected_array["frequency"]
+        )
+
+        pad_size = (p_len - len(f0) + 1) // 2
+        if pad_size > 0 or p_len - len(f0) - pad_size > 0:
+            f0 = np.pad(f0, [[pad_size, p_len - len(f0) - pad_size]], mode="constant")
+        f0, uv = self.interpolate_f0(f0)
+        return f0
+
+    def compute_f0_uv(self, wav, p_len=None):
+        x = wav
+        if p_len is None:
+            p_len = x.shape[0] // self.hop_length
+        else:
+            assert abs(p_len - x.shape[0] // self.hop_length) < 4, "pad length error"
+        time_step = self.hop_length / self.sampling_rate * 1000
+        f0 = (
+            parselmouth.Sound(x, self.sampling_rate)
+            .to_pitch_ac(
+                time_step=time_step / 1000,
+                voicing_threshold=0.6,
+                pitch_floor=self.f0_min,
+                pitch_ceiling=self.f0_max,
+            )
+            .selected_array["frequency"]
+        )
+
+        pad_size = (p_len - len(f0) + 1) // 2
+        if pad_size > 0 or p_len - len(f0) - pad_size > 0:
+            f0 = np.pad(f0, [[pad_size, p_len - len(f0) - pad_size]], mode="constant")
+        f0, uv = self.interpolate_f0(f0)
+        return f0, uv

lib/infer_pack/onnx_inference.py

@@ -0,0 +1,145 @@
+import onnxruntime
+import librosa
+import numpy as np
+import soundfile
+
+
+class ContentVec:
+    def __init__(self, vec_path="pretrained/vec-768-layer-12.onnx", device=None):
+        print("load model(s) from {}".format(vec_path))
+        if device == "cpu" or device is None:
+            providers = ["CPUExecutionProvider"]
+        elif device == "cuda":
+            providers = ["CUDAExecutionProvider", "CPUExecutionProvider"]
+        elif device == "dml":
+            providers = ["DmlExecutionProvider"]
+        else:
+            raise RuntimeError("Unsportted Device")
+        self.model = onnxruntime.InferenceSession(vec_path, providers=providers)
+
+    def __call__(self, wav):
+        return self.forward(wav)
+
+    def forward(self, wav):
+        feats = wav
+        if feats.ndim == 2:  # double channels
+            feats = feats.mean(-1)
+        assert feats.ndim == 1, feats.ndim
+        feats = np.expand_dims(np.expand_dims(feats, 0), 0)
+        onnx_input = {self.model.get_inputs()[0].name: feats}
+        logits = self.model.run(None, onnx_input)[0]
+        return logits.transpose(0, 2, 1)
+
+
+def get_f0_predictor(f0_predictor, hop_length, sampling_rate, **kargs):
+    if f0_predictor == "pm":
+        from lib.infer_pack.modules.F0Predictor.PMF0Predictor import PMF0Predictor
+
+        f0_predictor_object = PMF0Predictor(
+            hop_length=hop_length, sampling_rate=sampling_rate
+        )
+    elif f0_predictor == "harvest":
+        from lib.infer_pack.modules.F0Predictor.HarvestF0Predictor import (
+            HarvestF0Predictor,
+        )
+
+        f0_predictor_object = HarvestF0Predictor(
+            hop_length=hop_length, sampling_rate=sampling_rate
+        )
+    elif f0_predictor == "dio":
+        from lib.infer_pack.modules.F0Predictor.DioF0Predictor import DioF0Predictor
+
+        f0_predictor_object = DioF0Predictor(
+            hop_length=hop_length, sampling_rate=sampling_rate
+        )
+    else:
+        raise Exception("Unknown f0 predictor")
+    return f0_predictor_object
+
+
+class OnnxRVC:
+    def __init__(
+        self,
+        model_path,
+        sr=40000,
+        hop_size=512,
+        vec_path="vec-768-layer-12",
+        device="cpu",
+    ):
+        vec_path = f"pretrained/{vec_path}.onnx"
+        self.vec_model = ContentVec(vec_path, device)
+        if device == "cpu" or device is None:
+            providers = ["CPUExecutionProvider"]
+        elif device == "cuda":
+            providers = ["CUDAExecutionProvider", "CPUExecutionProvider"]
+        elif device == "dml":
+            providers = ["DmlExecutionProvider"]
+        else:
+            raise RuntimeError("Unsportted Device")
+        self.model = onnxruntime.InferenceSession(model_path, providers=providers)
+        self.sampling_rate = sr
+        self.hop_size = hop_size
+
+    def forward(self, hubert, hubert_length, pitch, pitchf, ds, rnd):
+        onnx_input = {
+            self.model.get_inputs()[0].name: hubert,
+            self.model.get_inputs()[1].name: hubert_length,
+            self.model.get_inputs()[2].name: pitch,
+            self.model.get_inputs()[3].name: pitchf,
+            self.model.get_inputs()[4].name: ds,
+            self.model.get_inputs()[5].name: rnd,
+        }
+        return (self.model.run(None, onnx_input)[0] * 32767).astype(np.int16)
+
+    def inference(
+        self,
+        raw_path,
+        sid,
+        f0_method="dio",
+        f0_up_key=0,
+        pad_time=0.5,
+        cr_threshold=0.02,
+    ):
+        f0_min = 50
+        f0_max = 1100
+        f0_mel_min = 1127 * np.log(1 + f0_min / 700)
+        f0_mel_max = 1127 * np.log(1 + f0_max / 700)
+        f0_predictor = get_f0_predictor(
+            f0_method,
+            hop_length=self.hop_size,
+            sampling_rate=self.sampling_rate,
+            threshold=cr_threshold,
+        )
+        wav, sr = librosa.load(raw_path, sr=self.sampling_rate)
+        org_length = len(wav)
+        if org_length / sr > 50.0:
+            raise RuntimeError("Reached Max Length")
+
+        wav16k = librosa.resample(wav, orig_sr=self.sampling_rate, target_sr=16000)
+        wav16k = wav16k
+
+        hubert = self.vec_model(wav16k)
+        hubert = np.repeat(hubert, 2, axis=2).transpose(0, 2, 1).astype(np.float32)
+        hubert_length = hubert.shape[1]
+
+        pitchf = f0_predictor.compute_f0(wav, hubert_length)
+        pitchf = pitchf * 2 ** (f0_up_key / 12)
+        pitch = pitchf.copy()
+        f0_mel = 1127 * np.log(1 + pitch / 700)
+        f0_mel[f0_mel > 0] = (f0_mel[f0_mel > 0] - f0_mel_min) * 254 / (
+            f0_mel_max - f0_mel_min
+        ) + 1
+        f0_mel[f0_mel <= 1] = 1
+        f0_mel[f0_mel > 255] = 255
+        pitch = np.rint(f0_mel).astype(np.int64)
+
+        pitchf = pitchf.reshape(1, len(pitchf)).astype(np.float32)
+        pitch = pitch.reshape(1, len(pitch))
+        ds = np.array([sid]).astype(np.int64)
+
+        rnd = np.random.randn(1, 192, hubert_length).astype(np.float32)
+        hubert_length = np.array([hubert_length]).astype(np.int64)
+
+        out_wav = self.forward(hubert, hubert_length, pitch, pitchf, ds, rnd).squeeze()
+        out_wav = np.pad(out_wav, (0, 2 * self.hop_size), "constant")
+        return out_wav[0:org_length]

lib/infer_pack/transforms.py

@@ -0,0 +1,209 @@
+import torch
+from torch.nn import functional as F
+
+import numpy as np
+
+
+DEFAULT_MIN_BIN_WIDTH = 1e-3
+DEFAULT_MIN_BIN_HEIGHT = 1e-3
+DEFAULT_MIN_DERIVATIVE = 1e-3
+
+
+def piecewise_rational_quadratic_transform(
+    inputs,
+    unnormalized_widths,
+    unnormalized_heights,
+    unnormalized_derivatives,
+    inverse=False,
+    tails=None,
+    tail_bound=1.0,
+    min_bin_width=DEFAULT_MIN_BIN_WIDTH,
+    min_bin_height=DEFAULT_MIN_BIN_HEIGHT,
+    min_derivative=DEFAULT_MIN_DERIVATIVE,
+):
+    if tails is None:
+        spline_fn = rational_quadratic_spline
+        spline_kwargs = {}
+    else:
+        spline_fn = unconstrained_rational_quadratic_spline
+        spline_kwargs = {"tails": tails, "tail_bound": tail_bound}
+
+    outputs, logabsdet = spline_fn(
+        inputs=inputs,
+        unnormalized_widths=unnormalized_widths,
+        unnormalized_heights=unnormalized_heights,
+        unnormalized_derivatives=unnormalized_derivatives,
+        inverse=inverse,
+        min_bin_width=min_bin_width,
+        min_bin_height=min_bin_height,
+        min_derivative=min_derivative,
+        **spline_kwargs
+    )
+    return outputs, logabsdet
+
+
+def searchsorted(bin_locations, inputs, eps=1e-6):
+    bin_locations[..., -1] += eps
+    return torch.sum(inputs[..., None] >= bin_locations, dim=-1) - 1
+
+
+def unconstrained_rational_quadratic_spline(
+    inputs,
+    unnormalized_widths,
+    unnormalized_heights,
+    unnormalized_derivatives,
+    inverse=False,
+    tails="linear",
+    tail_bound=1.0,
+    min_bin_width=DEFAULT_MIN_BIN_WIDTH,
+    min_bin_height=DEFAULT_MIN_BIN_HEIGHT,
+    min_derivative=DEFAULT_MIN_DERIVATIVE,
+):
+    inside_interval_mask = (inputs >= -tail_bound) & (inputs <= tail_bound)
+    outside_interval_mask = ~inside_interval_mask
+
+    outputs = torch.zeros_like(inputs)
+    logabsdet = torch.zeros_like(inputs)
+
+    if tails == "linear":
+        unnormalized_derivatives = F.pad(unnormalized_derivatives, pad=(1, 1))
+        constant = np.log(np.exp(1 - min_derivative) - 1)
+        unnormalized_derivatives[..., 0] = constant
+        unnormalized_derivatives[..., -1] = constant
+
+        outputs[outside_interval_mask] = inputs[outside_interval_mask]
+        logabsdet[outside_interval_mask] = 0
+    else:
+        raise RuntimeError("{} tails are not implemented.".format(tails))
+
+    (
+        outputs[inside_interval_mask],
+        logabsdet[inside_interval_mask],
+    ) = rational_quadratic_spline(
+        inputs=inputs[inside_interval_mask],
+        unnormalized_widths=unnormalized_widths[inside_interval_mask, :],
+        unnormalized_heights=unnormalized_heights[inside_interval_mask, :],
+        unnormalized_derivatives=unnormalized_derivatives[inside_interval_mask, :],
+        inverse=inverse,
+        left=-tail_bound,
+        right=tail_bound,
+        bottom=-tail_bound,
+        top=tail_bound,
+        min_bin_width=min_bin_width,
+        min_bin_height=min_bin_height,
+        min_derivative=min_derivative,
+    )
+
+    return outputs, logabsdet
+
+
+def rational_quadratic_spline(
+    inputs,
+    unnormalized_widths,
+    unnormalized_heights,
+    unnormalized_derivatives,
+    inverse=False,
+    left=0.0,
+    right=1.0,
+    bottom=0.0,
+    top=1.0,
+    min_bin_width=DEFAULT_MIN_BIN_WIDTH,
+    min_bin_height=DEFAULT_MIN_BIN_HEIGHT,
+    min_derivative=DEFAULT_MIN_DERIVATIVE,
+):
+    if torch.min(inputs) < left or torch.max(inputs) > right:
+        raise ValueError("Input to a transform is not within its domain")
+
+    num_bins = unnormalized_widths.shape[-1]
+
+    if min_bin_width * num_bins > 1.0:
+        raise ValueError("Minimal bin width too large for the number of bins")
+    if min_bin_height * num_bins > 1.0:
+        raise ValueError("Minimal bin height too large for the number of bins")
+
+    widths = F.softmax(unnormalized_widths, dim=-1)
+    widths = min_bin_width + (1 - min_bin_width * num_bins) * widths
+    cumwidths = torch.cumsum(widths, dim=-1)
+    cumwidths = F.pad(cumwidths, pad=(1, 0), mode="constant", value=0.0)
+    cumwidths = (right - left) * cumwidths + left
+    cumwidths[..., 0] = left
+    cumwidths[..., -1] = right
+    widths = cumwidths[..., 1:] - cumwidths[..., :-1]
+
+    derivatives = min_derivative + F.softplus(unnormalized_derivatives)
+
+    heights = F.softmax(unnormalized_heights, dim=-1)
+    heights = min_bin_height + (1 - min_bin_height * num_bins) * heights
+    cumheights = torch.cumsum(heights, dim=-1)
+    cumheights = F.pad(cumheights, pad=(1, 0), mode="constant", value=0.0)
+    cumheights = (top - bottom) * cumheights + bottom
+    cumheights[..., 0] = bottom
+    cumheights[..., -1] = top
+    heights = cumheights[..., 1:] - cumheights[..., :-1]
+
+    if inverse:
+        bin_idx = searchsorted(cumheights, inputs)[..., None]
+    else:
+        bin_idx = searchsorted(cumwidths, inputs)[..., None]
+
+    input_cumwidths = cumwidths.gather(-1, bin_idx)[..., 0]
+    input_bin_widths = widths.gather(-1, bin_idx)[..., 0]
+
+    input_cumheights = cumheights.gather(-1, bin_idx)[..., 0]
+    delta = heights / widths
+    input_delta = delta.gather(-1, bin_idx)[..., 0]
+
+    input_derivatives = derivatives.gather(-1, bin_idx)[..., 0]
+    input_derivatives_plus_one = derivatives[..., 1:].gather(-1, bin_idx)[..., 0]
+
+    input_heights = heights.gather(-1, bin_idx)[..., 0]
+
+    if inverse:
+        a = (inputs - input_cumheights) * (
+            input_derivatives + input_derivatives_plus_one - 2 * input_delta
+        ) + input_heights * (input_delta - input_derivatives)
+        b = input_heights * input_derivatives - (inputs - input_cumheights) * (
+            input_derivatives + input_derivatives_plus_one - 2 * input_delta
+        )
+        c = -input_delta * (inputs - input_cumheights)
+
+        discriminant = b.pow(2) - 4 * a * c
+        assert (discriminant >= 0).all()
+
+        root = (2 * c) / (-b - torch.sqrt(discriminant))
+        outputs = root * input_bin_widths + input_cumwidths
+
+        theta_one_minus_theta = root * (1 - root)
+        denominator = input_delta + (
+            (input_derivatives + input_derivatives_plus_one - 2 * input_delta)
+            * theta_one_minus_theta
+        )
+        derivative_numerator = input_delta.pow(2) * (
+            input_derivatives_plus_one * root.pow(2)
+            + 2 * input_delta * theta_one_minus_theta
+            + input_derivatives * (1 - root).pow(2)
+        )
+        logabsdet = torch.log(derivative_numerator) - 2 * torch.log(denominator)
+
+        return outputs, -logabsdet
+    else:
+        theta = (inputs - input_cumwidths) / input_bin_widths
+        theta_one_minus_theta = theta * (1 - theta)
+
+        numerator = input_heights * (
+            input_delta * theta.pow(2) + input_derivatives * theta_one_minus_theta
+        )
+        denominator = input_delta + (
+            (input_derivatives + input_derivatives_plus_one - 2 * input_delta)
+            * theta_one_minus_theta
+        )
+        outputs = input_cumheights + numerator / denominator
+
+        derivative_numerator = input_delta.pow(2) * (
+            input_derivatives_plus_one * theta.pow(2)
+            + 2 * input_delta * theta_one_minus_theta
+            + input_derivatives * (1 - theta).pow(2)
+        )
+        logabsdet = torch.log(derivative_numerator) - 2 * torch.log(denominator)
+
+        return outputs, logabsdet

lib/rmvpe.py

@@ -0,0 +1,432 @@
+import torch, numpy as np
+import torch.nn as nn
+import torch.nn.functional as F
+
+
+
+class BiGRU(nn.Module):
+    def __init__(self, input_features, hidden_features, num_layers):
+        super(BiGRU, self).__init__()
+        self.gru = nn.GRU(
+            input_features,
+            hidden_features,
+            num_layers=num_layers,
+            batch_first=True,
+            bidirectional=True,
+        )
+
+    def forward(self, x):
+        return self.gru(x)[0]
+
+
+class ConvBlockRes(nn.Module):
+    def __init__(self, in_channels, out_channels, momentum=0.01):
+        super(ConvBlockRes, self).__init__()
+        self.conv = nn.Sequential(
+            nn.Conv2d(
+                in_channels=in_channels,
+                out_channels=out_channels,
+                kernel_size=(3, 3),
+                stride=(1, 1),
+                padding=(1, 1),
+                bias=False,
+            ),
+            nn.BatchNorm2d(out_channels, momentum=momentum),
+            nn.ReLU(),
+            nn.Conv2d(
+                in_channels=out_channels,
+                out_channels=out_channels,
+                kernel_size=(3, 3),
+                stride=(1, 1),
+                padding=(1, 1),
+                bias=False,
+            ),
+            nn.BatchNorm2d(out_channels, momentum=momentum),
+            nn.ReLU(),
+        )
+        if in_channels != out_channels:
+            self.shortcut = nn.Conv2d(in_channels, out_channels, (1, 1))
+            self.is_shortcut = True
+        else:
+            self.is_shortcut = False
+
+    def forward(self, x):
+        if self.is_shortcut:
+            return self.conv(x) + self.shortcut(x)
+        else:
+            return self.conv(x) + x
+
+
+class Encoder(nn.Module):
+    def __init__(
+        self,
+        in_channels,
+        in_size,
+        n_encoders,
+        kernel_size,
+        n_blocks,
+        out_channels=16,
+        momentum=0.01,
+    ):
+        super(Encoder, self).__init__()
+        self.n_encoders = n_encoders
+        self.bn = nn.BatchNorm2d(in_channels, momentum=momentum)
+        self.layers = nn.ModuleList()
+        self.latent_channels = []
+        for i in range(self.n_encoders):
+            self.layers.append(
+                ResEncoderBlock(
+                    in_channels, out_channels, kernel_size, n_blocks, momentum=momentum
+                )
+            )
+            self.latent_channels.append([out_channels, in_size])
+            in_channels = out_channels
+            out_channels *= 2
+            in_size //= 2
+        self.out_size = in_size
+        self.out_channel = out_channels
+
+    def forward(self, x):
+        concat_tensors = []
+        x = self.bn(x)
+        for i in range(self.n_encoders):
+            _, x = self.layers[i](x)
+            concat_tensors.append(_)
+        return x, concat_tensors
+
+
+class ResEncoderBlock(nn.Module):
+    def __init__(
+        self, in_channels, out_channels, kernel_size, n_blocks=1, momentum=0.01
+    ):
+        super(ResEncoderBlock, self).__init__()
+        self.n_blocks = n_blocks
+        self.conv = nn.ModuleList()
+        self.conv.append(ConvBlockRes(in_channels, out_channels, momentum))
+        for i in range(n_blocks - 1):
+            self.conv.append(ConvBlockRes(out_channels, out_channels, momentum))
+        self.kernel_size = kernel_size
+        if self.kernel_size is not None:
+            self.pool = nn.AvgPool2d(kernel_size=kernel_size)
+
+    def forward(self, x):
+        for i in range(self.n_blocks):
+            x = self.conv[i](x)
+        if self.kernel_size is not None:
+            return x, self.pool(x)
+        else:
+            return x
+
+
+class Intermediate(nn.Module):  #
+    def __init__(self, in_channels, out_channels, n_inters, n_blocks, momentum=0.01):
+        super(Intermediate, self).__init__()
+        self.n_inters = n_inters
+        self.layers = nn.ModuleList()
+        self.layers.append(
+            ResEncoderBlock(in_channels, out_channels, None, n_blocks, momentum)
+        )
+        for i in range(self.n_inters - 1):
+            self.layers.append(
+                ResEncoderBlock(out_channels, out_channels, None, n_blocks, momentum)
+            )
+
+    def forward(self, x):
+        for i in range(self.n_inters):
+            x = self.layers[i](x)
+        return x
+
+
+class ResDecoderBlock(nn.Module):
+    def __init__(self, in_channels, out_channels, stride, n_blocks=1, momentum=0.01):
+        super(ResDecoderBlock, self).__init__()
+        out_padding = (0, 1) if stride == (1, 2) else (1, 1)
+        self.n_blocks = n_blocks
+        self.conv1 = nn.Sequential(
+            nn.ConvTranspose2d(
+                in_channels=in_channels,
+                out_channels=out_channels,
+                kernel_size=(3, 3),
+                stride=stride,
+                padding=(1, 1),
+                output_padding=out_padding,
+                bias=False,
+            ),
+            nn.BatchNorm2d(out_channels, momentum=momentum),
+            nn.ReLU(),
+        )
+        self.conv2 = nn.ModuleList()
+        self.conv2.append(ConvBlockRes(out_channels * 2, out_channels, momentum))
+        for i in range(n_blocks - 1):
+            self.conv2.append(ConvBlockRes(out_channels, out_channels, momentum))
+
+    def forward(self, x, concat_tensor):
+        x = self.conv1(x)
+        x = torch.cat((x, concat_tensor), dim=1)
+        for i in range(self.n_blocks):
+            x = self.conv2[i](x)
+        return x
+
+
+class Decoder(nn.Module):
+    def __init__(self, in_channels, n_decoders, stride, n_blocks, momentum=0.01):
+        super(Decoder, self).__init__()
+        self.layers = nn.ModuleList()
+        self.n_decoders = n_decoders
+        for i in range(self.n_decoders):
+            out_channels = in_channels // 2
+            self.layers.append(
+                ResDecoderBlock(in_channels, out_channels, stride, n_blocks, momentum)
+            )
+            in_channels = out_channels
+
+    def forward(self, x, concat_tensors):
+        for i in range(self.n_decoders):
+            x = self.layers[i](x, concat_tensors[-1 - i])
+        return x
+
+
+class DeepUnet(nn.Module):
+    def __init__(
+        self,
+        kernel_size,
+        n_blocks,
+        en_de_layers=5,
+        inter_layers=4,
+        in_channels=1,
+        en_out_channels=16,
+    ):
+        super(DeepUnet, self).__init__()
+        self.encoder = Encoder(
+            in_channels, 128, en_de_layers, kernel_size, n_blocks, en_out_channels
+        )
+        self.intermediate = Intermediate(
+            self.encoder.out_channel // 2,
+            self.encoder.out_channel,
+            inter_layers,
+            n_blocks,
+        )
+        self.decoder = Decoder(
+            self.encoder.out_channel, en_de_layers, kernel_size, n_blocks
+        )
+
+    def forward(self, x):
+        x, concat_tensors = self.encoder(x)
+        x = self.intermediate(x)
+        x = self.decoder(x, concat_tensors)
+        return x
+
+
+class E2E(nn.Module):
+    def __init__(
+        self,
+        n_blocks,
+        n_gru,
+        kernel_size,
+        en_de_layers=5,
+        inter_layers=4,
+        in_channels=1,
+        en_out_channels=16,
+    ):
+        super(E2E, self).__init__()
+        self.unet = DeepUnet(
+            kernel_size,
+            n_blocks,
+            en_de_layers,
+            inter_layers,
+            in_channels,
+            en_out_channels,
+        )
+        self.cnn = nn.Conv2d(en_out_channels, 3, (3, 3), padding=(1, 1))
+        if n_gru:
+            self.fc = nn.Sequential(
+                BiGRU(3 * 128, 256, n_gru),
+                nn.Linear(512, 360),
+                nn.Dropout(0.25),
+                nn.Sigmoid(),
+            )
+        else:
+            self.fc = nn.Sequential(
+                nn.Linear(3 * nn.N_MELS, nn.N_CLASS), nn.Dropout(0.25), nn.Sigmoid()
+            )
+
+    def forward(self, mel):
+        mel = mel.transpose(-1, -2).unsqueeze(1)
+        x = self.cnn(self.unet(mel)).transpose(1, 2).flatten(-2)
+        x = self.fc(x)
+        return x
+
+
+from librosa.filters import mel
+
+
+class MelSpectrogram(torch.nn.Module):
+    def __init__(
+        self,
+        is_half,
+        n_mel_channels,
+        sampling_rate,
+        win_length,
+        hop_length,
+        n_fft=None,
+        mel_fmin=0,
+        mel_fmax=None,
+        clamp=1e-5,
+    ):
+        super().__init__()
+        n_fft = win_length if n_fft is None else n_fft
+        self.hann_window = {}
+        mel_basis = mel(
+            sr=sampling_rate,
+            n_fft=n_fft,
+            n_mels=n_mel_channels,
+            fmin=mel_fmin,
+            fmax=mel_fmax,
+            htk=True,
+        )
+        mel_basis = torch.from_numpy(mel_basis).float()
+        self.register_buffer("mel_basis", mel_basis)
+        self.n_fft = win_length if n_fft is None else n_fft
+        self.hop_length = hop_length
+        self.win_length = win_length
+        self.sampling_rate = sampling_rate
+        self.n_mel_channels = n_mel_channels
+        self.clamp = clamp
+        self.is_half = is_half
+
+    def forward(self, audio, keyshift=0, speed=1, center=True):
+        factor = 2 ** (keyshift / 12)
+        n_fft_new = int(np.round(self.n_fft * factor))
+        win_length_new = int(np.round(self.win_length * factor))
+        hop_length_new = int(np.round(self.hop_length * speed))
+        keyshift_key = str(keyshift) + "_" + str(audio.device)
+        if keyshift_key not in self.hann_window:
+            self.hann_window[keyshift_key] = torch.hann_window(win_length_new).to(
+                audio.device
+            )
+        fft = torch.stft(
+            audio,
+            n_fft=n_fft_new,
+            hop_length=hop_length_new,
+            win_length=win_length_new,
+            window=self.hann_window[keyshift_key],
+            center=center,
+            return_complex=True,
+        )
+        magnitude = torch.sqrt(fft.real.pow(2) + fft.imag.pow(2))
+        if keyshift != 0:
+            size = self.n_fft // 2 + 1
+            resize = magnitude.size(1)
+            if resize < size:
+                magnitude = F.pad(magnitude, (0, 0, 0, size - resize))
+            magnitude = magnitude[:, :size, :] * self.win_length / win_length_new
+        mel_output = torch.matmul(self.mel_basis, magnitude)
+        if self.is_half == True:
+            mel_output = mel_output.half()
+        log_mel_spec = torch.log(torch.clamp(mel_output, min=self.clamp))
+        return log_mel_spec
+
+
+class RMVPE:
+    def __init__(self, model_path, is_half, device=None):
+        self.resample_kernel = {}
+        model = E2E(4, 1, (2, 2))
+        ckpt = torch.load(model_path, map_location="cpu")
+        model.load_state_dict(ckpt)
+        model.eval()
+        if is_half == True:
+            model = model.half()
+        self.model = model
+        self.resample_kernel = {}
+        self.is_half = is_half
+        if device is None:
+            device = "cuda" if torch.cuda.is_available() else "cpu"
+        self.device = device
+        self.mel_extractor = MelSpectrogram(
+            is_half, 128, 16000, 1024, 160, None, 30, 8000
+        ).to(device)
+        self.model = self.model.to(device)
+        cents_mapping = 20 * np.arange(360) + 1997.3794084376191
+        self.cents_mapping = np.pad(cents_mapping, (4, 4))  # 368
+
+    def mel2hidden(self, mel):
+        with torch.no_grad():
+            n_frames = mel.shape[-1]
+            mel = F.pad(
+                mel, (0, 32 * ((n_frames - 1) // 32 + 1) - n_frames), mode="reflect"
+            )
+            hidden = self.model(mel)
+            return hidden[:, :n_frames]
+
+    def decode(self, hidden, thred=0.03):
+        cents_pred = self.to_local_average_cents(hidden, thred=thred)
+        f0 = 10 * (2 ** (cents_pred / 1200))
+        f0[f0 == 10] = 0
+        # f0 = np.array([10 * (2 ** (cent_pred / 1200)) if cent_pred else 0 for cent_pred in cents_pred])
+        return f0
+
+    def infer_from_audio(self, audio, thred=0.03):
+        audio = torch.from_numpy(audio).float().to(self.device).unsqueeze(0)
+        # torch.cuda.synchronize()
+        # t0=ttime()
+        mel = self.mel_extractor(audio, center=True)
+        # torch.cuda.synchronize()
+        # t1=ttime()
+        hidden = self.mel2hidden(mel)
+        # torch.cuda.synchronize()
+        # t2=ttime()
+        hidden = hidden.squeeze(0).cpu().numpy()
+        if self.is_half == True:
+            hidden = hidden.astype("float32")
+        f0 = self.decode(hidden, thred=thred)
+        # torch.cuda.synchronize()
+        # t3=ttime()
+        # print("hmvpe:%s\t%s\t%s\t%s"%(t1-t0,t2-t1,t3-t2,t3-t0))
+        return f0
+
+    def to_local_average_cents(self, salience, thred=0.05):
+        # t0 = ttime()
+        center = np.argmax(salience, axis=1)  # frame length#index
+        salience = np.pad(salience, ((0, 0), (4, 4)))  # frame length,368
+        # t1 = ttime()
+        center += 4
+        todo_salience = []
+        todo_cents_mapping = []
+        starts = center - 4
+        ends = center + 5
+        for idx in range(salience.shape[0]):
+            todo_salience.append(salience[:, starts[idx] : ends[idx]][idx])
+            todo_cents_mapping.append(self.cents_mapping[starts[idx] : ends[idx]])
+        # t2 = ttime()
+        todo_salience = np.array(todo_salience)  # frame length，9
+        todo_cents_mapping = np.array(todo_cents_mapping)  # frame length，9
+        product_sum = np.sum(todo_salience * todo_cents_mapping, 1)
+        weight_sum = np.sum(todo_salience, 1)  # frame length
+        devided = product_sum / weight_sum  # frame length
+        # t3 = ttime()
+        maxx = np.max(salience, axis=1)  # frame length
+        devided[maxx <= thred] = 0
+        # t4 = ttime()
+        # print("decode:%s\t%s\t%s\t%s" % (t1 - t0, t2 - t1, t3 - t2, t4 - t3))
+        return devided
+
+
+# if __name__ == '__main__':
+#     audio, sampling_rate = sf.read("Quotations~1.wav") ### edit
+#     if len(audio.shape) > 1:
+#         audio = librosa.to_mono(audio.transpose(1, 0))
+#     audio_bak = audio.copy()
+#     if sampling_rate != 16000:
+#         audio = librosa.resample(audio, orig_sr=sampling_rate, target_sr=16000)
+#     model_path = "/bili-coeus/jupyter/jupyterhub-liujing04/vits_ch/test-RMVPE/weights/rmvpe_llc_half.pt"
+#     thred = 0.03  # 0.01
+#     device = 'cuda' if torch.cuda.is_available() else 'cpu'
+#     rmvpe = RMVPE(model_path,is_half=False, device=device)
+#     t0=ttime()
+#     f0 = rmvpe.infer_from_audio(audio, thred=thred)
+#     f0 = rmvpe.infer_from_audio(audio, thred=thred)
+#     f0 = rmvpe.infer_from_audio(audio, thred=thred)
+#     f0 = rmvpe.infer_from_audio(audio, thred=thred)
+#     f0 = rmvpe.infer_from_audio(audio, thred=thred)
+#     t1=ttime()
+#     print(f0.shape,t1-t0)

requirements_colab.txt

@@ -0,0 +1,9 @@
+--extra-index-url https://download.pytorch.org/whl/cu117 torch==2.0.0+cu117 torchvision==0.15.2+cu117 torchaudio==2.0.0
+yt-dlp 
+gradio==3.35.2
+pydub==0.25.1
+edge_tts==6.1.7
+deep_translator==1.11.4
+git+https://github.com/m-bain/whisperx.git
+gTTS
+gradio_client==0.2.7

requirements_extra.txt

@@ -0,0 +1,8 @@
+praat-parselmouth>=0.4.3
+pyworld==0.3.2
+faiss-cpu==1.7.3
+torchcrepe==0.0.20
+ffmpeg-python>=0.2.0
+fairseq==0.12.2 
+gdown
+rarfile

soni_translate/audio_segments.py

@@ -24,4 +24,4 @@ def create_translated_audio(result_diarize, audio_files, Output_name_file):
     os.system("rm -rf audio/*")
 
     # combined audio as a file
-    combined_audio.export(Output_name_file, format="wav")
+    combined_audio.export(Output_name_file, format="wav") # best than ogg, change if the audio is anomalous

soni_translate/text_to_speech.py

@@ -12,7 +12,7 @@ def make_voice(tts_text, tts_voice, filename,language):
       try:
           tts = gTTS(tts_text, lang=language)
           tts.save(filename)
-          print(f'No audio was received. Please change the tts voice for {tts_voice}. USING gTTS.')
+          print(f'No audio was received. Please change the tts voice for {tts_voice}. TTS auxiliary will be used in the segment')
       except:
         tts = gTTS('a', lang=language)
         tts.save(filename)
@@ -26,7 +26,7 @@ def make_voice_gradio(tts_text, tts_voice, filename, language):
       try:
         tts = gTTS(tts_text, lang=language)
         tts.save(filename)
-        print(f'No audio was received. Please change the tts voice for {tts_voice}. USING gTTS.')
+        print(f'No audio was received. Please change the tts voice for {tts_voice}. TTS auxiliary will be used in the segment')
       except:
         tts = gTTS('a', lang=language)
         tts.save(filename)

soni_translate/translate_segments.py

@@ -2,9 +2,6 @@ from tqdm import tqdm
 from deep_translator import GoogleTranslator
 
 def translate_text(segments, TRANSLATE_AUDIO_TO):
-
-    if TRANSLATE_AUDIO_TO == "zh":
-        TRANSLATE_AUDIO_TO = "zh-CN"
     
     translator = GoogleTranslator(source='auto', target=TRANSLATE_AUDIO_TO)
     

vc_infer_pipeline.py

@@ -0,0 +1,445 @@
+import numpy as np, parselmouth, torch, pdb, sys, os
+from time import time as ttime
+import torch.nn.functional as F
+import scipy.signal as signal
+import pyworld, os, traceback, faiss, librosa, torchcrepe
+from scipy import signal
+from functools import lru_cache
+
+now_dir = os.getcwd()
+sys.path.append(now_dir)
+
+bh, ah = signal.butter(N=5, Wn=48, btype="high", fs=16000)
+
+input_audio_path2wav = {}
+
+
+@lru_cache
+def cache_harvest_f0(input_audio_path, fs, f0max, f0min, frame_period):
+    audio = input_audio_path2wav[input_audio_path]
+    f0, t = pyworld.harvest(
+        audio,
+        fs=fs,
+        f0_ceil=f0max,
+        f0_floor=f0min,
+        frame_period=frame_period,
+    )
+    f0 = pyworld.stonemask(audio, f0, t, fs)
+    return f0
+
+
+def change_rms(data1, sr1, data2, sr2, rate):  # 1 is the input audio, 2 is the output audio, rate is the proportion of 2
+    # print(data1.max(),data2.max())
+    rms1 = librosa.feature.rms(
+        y=data1, frame_length=sr1 // 2 * 2, hop_length=sr1 // 2
+    )  # one dot every half second
+    rms2 = librosa.feature.rms(y=data2, frame_length=sr2 // 2 * 2, hop_length=sr2 // 2)
+    rms1 = torch.from_numpy(rms1)
+    rms1 = F.interpolate(
+        rms1.unsqueeze(0), size=data2.shape[0], mode="linear"
+    ).squeeze()
+    rms2 = torch.from_numpy(rms2)
+    rms2 = F.interpolate(
+        rms2.unsqueeze(0), size=data2.shape[0], mode="linear"
+    ).squeeze()
+    rms2 = torch.max(rms2, torch.zeros_like(rms2) + 1e-6)
+    data2 *= (
+        torch.pow(rms1, torch.tensor(1 - rate))
+        * torch.pow(rms2, torch.tensor(rate - 1))
+    ).numpy()
+    return data2
+
+
+class VC(object):
+    def __init__(self, tgt_sr, config):
+        self.x_pad, self.x_query, self.x_center, self.x_max, self.is_half = (
+            config.x_pad,
+            config.x_query,
+            config.x_center,
+            config.x_max,
+            config.is_half,
+        )
+        self.sr = 16000  # hubert input sampling rate
+        self.window = 160  # points per frame
+        self.t_pad = self.sr * self.x_pad  # Pad time before and after each bar
+        self.t_pad_tgt = tgt_sr * self.x_pad
+        self.t_pad2 = self.t_pad * 2
+        self.t_query = self.sr * self.x_query  # Query time before and after the cut point
+        self.t_center = self.sr * self.x_center  # Query point cut position
+        self.t_max = self.sr * self.x_max  # Query-free duration threshold
+        self.device = config.device
+
+    def get_f0(
+        self,
+        input_audio_path,
+        x,
+        p_len,
+        f0_up_key,
+        f0_method,
+        filter_radius,
+        inp_f0=None,
+    ):
+        global input_audio_path2wav
+        time_step = self.window / self.sr * 1000
+        f0_min = 50
+        f0_max = 1100
+        f0_mel_min = 1127 * np.log(1 + f0_min / 700)
+        f0_mel_max = 1127 * np.log(1 + f0_max / 700)
+        if f0_method == "pm":
+            f0 = (
+                parselmouth.Sound(x, self.sr)
+                .to_pitch_ac(
+                    time_step=time_step / 1000,
+                    voicing_threshold=0.6,
+                    pitch_floor=f0_min,
+                    pitch_ceiling=f0_max,
+                )
+                .selected_array["frequency"]
+            )
+            pad_size = (p_len - len(f0) + 1) // 2
+            if pad_size > 0 or p_len - len(f0) - pad_size > 0:
+                f0 = np.pad(
+                    f0, [[pad_size, p_len - len(f0) - pad_size]], mode="constant"
+                )
+        elif f0_method == "harvest":
+            input_audio_path2wav[input_audio_path] = x.astype(np.double)
+            f0 = cache_harvest_f0(input_audio_path, self.sr, f0_max, f0_min, 10)
+            if filter_radius > 2:
+                f0 = signal.medfilt(f0, 3)
+        elif f0_method == "crepe":
+            model = "full"
+            # Pick a batch size that doesn't cause memory errors on your gpu
+            batch_size = 512
+            # Compute pitch using first gpu
+            audio = torch.tensor(np.copy(x))[None].float()
+            f0, pd = torchcrepe.predict(
+                audio,
+                self.sr,
+                self.window,
+                f0_min,
+                f0_max,
+                model,
+                batch_size=batch_size,
+                device=self.device,
+                return_periodicity=True,
+            )
+            pd = torchcrepe.filter.median(pd, 3)
+            f0 = torchcrepe.filter.mean(f0, 3)
+            f0[pd < 0.1] = 0
+            f0 = f0[0].cpu().numpy()
+        elif f0_method == "rmvpe":
+            if hasattr(self, "model_rmvpe") == False:
+                from lib.rmvpe import RMVPE
+
+                print("loading rmvpe model")
+                self.model_rmvpe = RMVPE(
+                    "rmvpe.pt", is_half=self.is_half, device=self.device
+                )
+            f0 = self.model_rmvpe.infer_from_audio(x, thred=0.03)
+        f0 *= pow(2, f0_up_key / 12)
+        # with open("test.txt","w")as f:f.write("\n".join([str(i)for i in f0.tolist()]))
+        tf0 = self.sr // self.window  # f0 points per second
+        if inp_f0 is not None:
+            delta_t = np.round(
+                (inp_f0[:, 0].max() - inp_f0[:, 0].min()) * tf0 + 1
+            ).astype("int16")
+            replace_f0 = np.interp(
+                list(range(delta_t)), inp_f0[:, 0] * 100, inp_f0[:, 1]
+            )
+            shape = f0[self.x_pad * tf0 : self.x_pad * tf0 + len(replace_f0)].shape[0]
+            f0[self.x_pad * tf0 : self.x_pad * tf0 + len(replace_f0)] = replace_f0[
+                :shape
+            ]
+        # with open("test_opt.txt","w")as f:f.write("\n".join([str(i)for i in f0.tolist()]))
+        f0bak = f0.copy()
+        f0_mel = 1127 * np.log(1 + f0 / 700)
+        f0_mel[f0_mel > 0] = (f0_mel[f0_mel > 0] - f0_mel_min) * 254 / (
+            f0_mel_max - f0_mel_min
+        ) + 1
+        f0_mel[f0_mel <= 1] = 1
+        f0_mel[f0_mel > 255] = 255
+        f0_coarse = np.rint(f0_mel).astype(np.int)
+        return f0_coarse, f0bak  # 1-0
+
+    def vc(
+        self,
+        model,
+        net_g,
+        sid,
+        audio0,
+        pitch,
+        pitchf,
+        times,
+        index,
+        big_npy,
+        index_rate,
+        version,
+        protect,
+    ):  # ,file_index,file_big_npy
+        feats = torch.from_numpy(audio0)
+        if self.is_half:
+            feats = feats.half()
+        else:
+            feats = feats.float()
+        if feats.dim() == 2:  # double channels
+            feats = feats.mean(-1)
+        assert feats.dim() == 1, feats.dim()
+        feats = feats.view(1, -1)
+        padding_mask = torch.BoolTensor(feats.shape).to(self.device).fill_(False)
+
+        inputs = {
+            "source": feats.to(self.device),
+            "padding_mask": padding_mask,
+            "output_layer": 9 if version == "v1" else 12,
+        }
+        t0 = ttime()
+        with torch.no_grad():
+            logits = model.extract_features(**inputs)
+            feats = model.final_proj(logits[0]) if version == "v1" else logits[0]
+        if protect < 0.5 and pitch != None and pitchf != None:
+            feats0 = feats.clone()
+        if (
+            isinstance(index, type(None)) == False
+            and isinstance(big_npy, type(None)) == False
+            and index_rate != 0
+        ):
+            npy = feats[0].cpu().numpy()
+            if self.is_half:
+                npy = npy.astype("float32")
+
+            # _, I = index.search(npy, 1)
+            # npy = big_npy[I.squeeze()]
+
+            score, ix = index.search(npy, k=8)
+            weight = np.square(1 / score)
+            weight /= weight.sum(axis=1, keepdims=True)
+            npy = np.sum(big_npy[ix] * np.expand_dims(weight, axis=2), axis=1)
+
+            if self.is_half:
+                npy = npy.astype("float16")
+            feats = (
+                torch.from_numpy(npy).unsqueeze(0).to(self.device) * index_rate
+                + (1 - index_rate) * feats
+            )
+
+        feats = F.interpolate(feats.permute(0, 2, 1), scale_factor=2).permute(0, 2, 1)
+        if protect < 0.5 and pitch != None and pitchf != None:
+            feats0 = F.interpolate(feats0.permute(0, 2, 1), scale_factor=2).permute(
+                0, 2, 1
+            )
+        t1 = ttime()
+        p_len = audio0.shape[0] // self.window
+        if feats.shape[1] < p_len:
+            p_len = feats.shape[1]
+            if pitch != None and pitchf != None:
+                pitch = pitch[:, :p_len]
+                pitchf = pitchf[:, :p_len]
+
+        if protect < 0.5 and pitch != None and pitchf != None:
+            pitchff = pitchf.clone()
+            pitchff[pitchf > 0] = 1
+            pitchff[pitchf < 1] = protect
+            pitchff = pitchff.unsqueeze(-1)
+            feats = feats * pitchff + feats0 * (1 - pitchff)
+            feats = feats.to(feats0.dtype)
+        p_len = torch.tensor([p_len], device=self.device).long()
+        with torch.no_grad():
+            if pitch != None and pitchf != None:
+                audio1 = (
+                    (net_g.infer(feats, p_len, pitch, pitchf, sid)[0][0, 0])
+                    .data.cpu()
+                    .float()
+                    .numpy()
+                )
+            else:
+                audio1 = (
+                    (net_g.infer(feats, p_len, sid)[0][0, 0]).data.cpu().float().numpy()
+                )
+        del feats, p_len, padding_mask
+        if torch.cuda.is_available():
+            torch.cuda.empty_cache()
+        t2 = ttime()
+        times[0] += t1 - t0
+        times[2] += t2 - t1
+        return audio1
+
+    def pipeline(
+        self,
+        model,
+        net_g,
+        sid,
+        audio,
+        input_audio_path,
+        times,
+        f0_up_key,
+        f0_method,
+        file_index,
+        # file_big_npy,
+        index_rate,
+        if_f0,
+        filter_radius,
+        tgt_sr,
+        resample_sr,
+        rms_mix_rate,
+        version,
+        protect,
+        f0_file=None,
+    ):
+        if (
+            file_index != ""
+            # and file_big_npy != ""
+            # and os.path.exists(file_big_npy) == True
+            and os.path.exists(file_index) == True
+            and index_rate != 0
+        ):
+            try:
+                index = faiss.read_index(file_index)
+                # big_npy = np.load(file_big_npy)
+                big_npy = index.reconstruct_n(0, index.ntotal)
+            except:
+                traceback.print_exc()
+                index = big_npy = None
+        else:
+            index = big_npy = None
+            print("File index Not found, set None")
+
+        audio = signal.filtfilt(bh, ah, audio)
+        audio_pad = np.pad(audio, (self.window // 2, self.window // 2), mode="reflect")
+        opt_ts = []
+        if audio_pad.shape[0] > self.t_max:
+            audio_sum = np.zeros_like(audio)
+            for i in range(self.window):
+                audio_sum += audio_pad[i : i - self.window]
+            for t in range(self.t_center, audio.shape[0], self.t_center):
+                opt_ts.append(
+                    t
+                    - self.t_query
+                    + np.where(
+                        np.abs(audio_sum[t - self.t_query : t + self.t_query])
+                        == np.abs(audio_sum[t - self.t_query : t + self.t_query]).min()
+                    )[0][0]
+                )
+        s = 0
+        audio_opt = []
+        t = None
+        t1 = ttime()
+        audio_pad = np.pad(audio, (self.t_pad, self.t_pad), mode="reflect")
+        p_len = audio_pad.shape[0] // self.window
+        inp_f0 = None
+        if hasattr(f0_file, "name") == True:
+            try:
+                with open(f0_file.name, "r") as f:
+                    lines = f.read().strip("\n").split("\n")
+                inp_f0 = []
+                for line in lines:
+                    inp_f0.append([float(i) for i in line.split(",")])
+                inp_f0 = np.array(inp_f0, dtype="float32")
+            except:
+                traceback.print_exc()
+        sid = torch.tensor(sid, device=self.device).unsqueeze(0).long()
+        pitch, pitchf = None, None
+        if if_f0 == 1:
+            pitch, pitchf = self.get_f0(
+                input_audio_path,
+                audio_pad,
+                p_len,
+                f0_up_key,
+                f0_method,
+                filter_radius,
+                inp_f0,
+            )
+            pitch = pitch[:p_len]
+            pitchf = pitchf[:p_len]
+            if self.device == "mps":
+                pitchf = pitchf.astype(np.float32)
+            pitch = torch.tensor(pitch, device=self.device).unsqueeze(0).long()
+            pitchf = torch.tensor(pitchf, device=self.device).unsqueeze(0).float()
+        t2 = ttime()
+        times[1] += t2 - t1
+        for t in opt_ts:
+            t = t // self.window * self.window
+            if if_f0 == 1:
+                audio_opt.append(
+                    self.vc(
+                        model,
+                        net_g,
+                        sid,
+                        audio_pad[s : t + self.t_pad2 + self.window],
+                        pitch[:, s // self.window : (t + self.t_pad2) // self.window],
+                        pitchf[:, s // self.window : (t + self.t_pad2) // self.window],
+                        times,
+                        index,
+                        big_npy,
+                        index_rate,
+                        version,
+                        protect,
+                    )[self.t_pad_tgt : -self.t_pad_tgt]
+                )
+            else:
+                audio_opt.append(
+                    self.vc(
+                        model,
+                        net_g,
+                        sid,
+                        audio_pad[s : t + self.t_pad2 + self.window],
+                        None,
+                        None,
+                        times,
+                        index,
+                        big_npy,
+                        index_rate,
+                        version,
+                        protect,
+                    )[self.t_pad_tgt : -self.t_pad_tgt]
+                )
+            s = t
+        if if_f0 == 1:
+            audio_opt.append(
+                self.vc(
+                    model,
+                    net_g,
+                    sid,
+                    audio_pad[t:],
+                    pitch[:, t // self.window :] if t is not None else pitch,
+                    pitchf[:, t // self.window :] if t is not None else pitchf,
+                    times,
+                    index,
+                    big_npy,
+                    index_rate,
+                    version,
+                    protect,
+                )[self.t_pad_tgt : -self.t_pad_tgt]
+            )
+        else:
+            audio_opt.append(
+                self.vc(
+                    model,
+                    net_g,
+                    sid,
+                    audio_pad[t:],
+                    None,
+                    None,
+                    times,
+                    index,
+                    big_npy,
+                    index_rate,
+                    version,
+                    protect,
+                )[self.t_pad_tgt : -self.t_pad_tgt]
+            )
+        audio_opt = np.concatenate(audio_opt)
+        if rms_mix_rate != 1:
+            audio_opt = change_rms(audio, 16000, audio_opt, tgt_sr, rms_mix_rate)
+        if resample_sr >= 16000 and tgt_sr != resample_sr:
+            audio_opt = librosa.resample(
+                audio_opt, orig_sr=tgt_sr, target_sr=resample_sr
+            )
+        audio_max = np.abs(audio_opt).max() / 0.99
+        max_int16 = 32768
+        if audio_max > 1:
+            max_int16 /= audio_max
+        audio_opt = (audio_opt * max_int16).astype(np.int16)
+        del pitch, pitchf, sid
+        if torch.cuda.is_available():
+            torch.cuda.empty_cache()
+        return audio_opt

voice_main.py

@@ -0,0 +1,554 @@
+import torch
+from lib.infer_pack.models import (
+    SynthesizerTrnMs256NSFsid,
+    SynthesizerTrnMs256NSFsid_nono,
+    SynthesizerTrnMs768NSFsid,
+    SynthesizerTrnMs768NSFsid_nono,
+)
+from vc_infer_pipeline import VC
+import traceback, pdb
+from lib.audio import load_audio
+import numpy as np
+import os
+from fairseq import checkpoint_utils
+import soundfile as sf
+from gtts import gTTS
+import edge_tts
+import asyncio
+import nest_asyncio
+
+# model load
+def get_vc(sid, to_return_protect0, to_return_protect1):
+    global n_spk, tgt_sr, net_g, vc, cpt, version
+    if sid == "" or sid == []:
+        global hubert_model
+        if hubert_model is not None:  # change model or not
+            print("clean_empty_cache")
+            del net_g, n_spk, vc, hubert_model, tgt_sr  # ,cpt
+            hubert_model = net_g = n_spk = vc = hubert_model = tgt_sr = None
+            if torch.cuda.is_available():
+                torch.cuda.empty_cache()
+            ### if clean
+            if_f0 = cpt.get("f0", 1)
+            version = cpt.get("version", "v1")
+            if version == "v1":
+                if if_f0 == 1:
+                    net_g = SynthesizerTrnMs256NSFsid(
+                        *cpt["config"], is_half=config.is_half
+                    )
+                else:
+                    net_g = SynthesizerTrnMs256NSFsid_nono(*cpt["config"])
+            elif version == "v2":
+                if if_f0 == 1:
+                    net_g = SynthesizerTrnMs768NSFsid(
+                        *cpt["config"], is_half=config.is_half
+                    )
+                else:
+                    net_g = SynthesizerTrnMs768NSFsid_nono(*cpt["config"])
+            del net_g, cpt
+            if torch.cuda.is_available():
+                torch.cuda.empty_cache()
+        return {"visible": False, "__type__": "update"}
+    person = "%s/%s" % (weight_root, sid)
+    print("loading %s" % person)
+    cpt = torch.load(person, map_location="cpu")
+    tgt_sr = cpt["config"][-1]
+    cpt["config"][-3] = cpt["weight"]["emb_g.weight"].shape[0]  # n_spk
+    if_f0 = cpt.get("f0", 1)
+    if if_f0 == 0:
+        to_return_protect0 = to_return_protect1 = {
+            "visible": False,
+            "value": 0.5,
+            "__type__": "update",
+        }
+    else:
+        to_return_protect0 = {
+            "visible": True,
+            "value": to_return_protect0,
+            "__type__": "update",
+        }
+        to_return_protect1 = {
+            "visible": True,
+            "value": to_return_protect1,
+            "__type__": "update",
+        }
+    version = cpt.get("version", "v1")
+    if version == "v1":
+        if if_f0 == 1:
+            net_g = SynthesizerTrnMs256NSFsid(*cpt["config"], is_half=config.is_half)
+        else:
+            net_g = SynthesizerTrnMs256NSFsid_nono(*cpt["config"])
+    elif version == "v2":
+        if if_f0 == 1:
+            net_g = SynthesizerTrnMs768NSFsid(*cpt["config"], is_half=config.is_half)
+        else:
+            net_g = SynthesizerTrnMs768NSFsid_nono(*cpt["config"])
+    del net_g.enc_q
+    print(net_g.load_state_dict(cpt["weight"], strict=False))
+    net_g.eval().to(config.device)
+    if config.is_half:
+        net_g = net_g.half()
+    else:
+        net_g = net_g.float()
+    vc = VC(tgt_sr, config)
+    n_spk = cpt["config"][-3]
+    return (
+        {"visible": True, "maximum": n_spk, "__type__": "update"},
+        to_return_protect0,
+        to_return_protect1,
+    )
+
+
+
+# inference
+def vc_single(
+    sid,
+    input_audio_path,
+    f0_up_key,
+    f0_file,
+    f0_method,
+    file_index,
+    file_index2,
+    # file_big_npy,
+    index_rate,
+    filter_radius,
+    resample_sr,
+    rms_mix_rate,
+    protect,
+):  
+    global tgt_sr, net_g, vc, hubert_model, version, cpt
+    if input_audio_path is None:
+        return "You need to upload an audio", None
+    f0_up_key = int(f0_up_key)
+    try:
+        audio = load_audio(input_audio_path, 16000)
+        audio_max = np.abs(audio).max() / 0.95
+        if audio_max > 1:
+            audio /= audio_max
+        times = [0, 0, 0]
+        if not hubert_model:
+            load_hubert()
+        if_f0 = cpt.get("f0", 1)
+        file_index = (
+            (
+                file_index.strip(" ")
+                .strip('"')
+                .strip("\n")
+                .strip('"')
+                .strip(" ")
+                .replace("trained", "added")
+            )
+            if file_index != ""
+            else file_index2
+        )  # reemplace for 2
+        # file_big_npy = (
+        #     file_big_npy.strip(" ").strip('"').strip("\n").strip('"').strip(" ")
+        # )
+        audio_opt = vc.pipeline(
+            hubert_model,
+            net_g,
+            sid,
+            audio,
+            input_audio_path,
+            times,
+            f0_up_key,
+            f0_method,
+            file_index,
+            # file_big_npy,
+            index_rate,
+            if_f0,
+            filter_radius,
+            tgt_sr,
+            resample_sr,
+            rms_mix_rate,
+            version,
+            protect,
+            f0_file=f0_file,
+        )
+        if tgt_sr != resample_sr >= 16000:
+            tgt_sr = resample_sr
+        index_info = (
+            "Using index:%s." % file_index
+            if os.path.exists(file_index)
+            else "Index not used."
+        )
+        return "Success.\n %s\nTime:\n npy:%ss, f0:%ss, infer:%ss" % (
+            index_info,
+            times[0],
+            times[1],
+            times[2],
+        ), (tgt_sr, audio_opt)
+    except:
+        info = traceback.format_exc()
+        print(info)
+        return info, (None, None)
+
+
+
+# hubert model
+def load_hubert():
+    global hubert_model
+    models, _, _ = checkpoint_utils.load_model_ensemble_and_task(
+        ["hubert_base.pt"],
+        suffix="",
+    )
+    hubert_model = models[0]
+    hubert_model = hubert_model.to(config.device)
+    if config.is_half:
+        hubert_model = hubert_model.half()
+    else:
+        hubert_model = hubert_model.float()
+    hubert_model.eval()
+
+# config cpu
+def use_fp32_config():
+    for config_file in [
+        "32k.json",
+        "40k.json",
+        "48k.json",
+        "48k_v2.json",
+        "32k_v2.json",
+    ]:
+        with open(f"configs/{config_file}", "r") as f:
+            strr = f.read().replace("true", "false")
+        with open(f"configs/{config_file}", "w") as f:
+            f.write(strr)
+
+# config device and torch type
+class Config:
+    def __init__(self, device, is_half):
+        self.device = device
+        self.is_half = is_half
+        self.n_cpu = 2 # set cpu cores ####################
+        self.gpu_name = None
+        self.gpu_mem = None
+        self.x_pad, self.x_query, self.x_center, self.x_max = self.device_config()
+
+    def device_config(self) -> tuple:
+        if torch.cuda.is_available():
+            i_device = int(self.device.split(":")[-1])
+            self.gpu_name = torch.cuda.get_device_name(i_device)
+            if (
+                ("16" in self.gpu_name and "V100" not in self.gpu_name.upper())
+                or "P40" in self.gpu_name.upper()
+                or "1060" in self.gpu_name
+                or "1070" in self.gpu_name
+                or "1080" in self.gpu_name
+            ):
+                print("16 series / 10 series graphics cards and P40 force single precision")
+                self.is_half = False
+                for config_file in ["32k.json", "40k.json", "48k.json"]:
+                    with open(f"configs/{config_file}", "r") as f:
+                        strr = f.read().replace("true", "false")
+                    with open(f"configs/{config_file}", "w") as f:
+                        f.write(strr)
+                with open("trainset_preprocess_pipeline_print.py", "r") as f:
+                    strr = f.read().replace("3.7", "3.0")
+                with open("trainset_preprocess_pipeline_print.py", "w") as f:
+                    f.write(strr)
+            else:
+                self.gpu_name = None
+            self.gpu_mem = int(
+                torch.cuda.get_device_properties(i_device).total_memory
+                / 1024
+                / 1024
+                / 1024
+                + 0.4
+            )
+            if self.gpu_mem <= 4:
+                with open("trainset_preprocess_pipeline_print.py", "r") as f:
+                    strr = f.read().replace("3.7", "3.0")
+                with open("trainset_preprocess_pipeline_print.py", "w") as f:
+                    f.write(strr)
+        elif torch.backends.mps.is_available():
+            print("Supported N-card not found, using MPS for inference")
+            self.device = "mps"
+        else:
+            print("No supported N-card found, using CPU for inference")
+            self.device = "cpu"
+            self.is_half = False
+            use_fp32_config()
+
+        if self.n_cpu == 0:
+            self.n_cpu = cpu_count()
+
+        if self.is_half:
+            # 6GB VRAM configuration
+            x_pad = 3
+            x_query = 10
+            x_center = 60
+            x_max = 65
+        else:
+            # 5GB VRAM configuration
+            x_pad = 1
+            x_query = 6
+            x_center = 38
+            x_max = 41
+
+        if self.gpu_mem != None and self.gpu_mem <= 4:
+            x_pad = 1
+            x_query = 5
+            x_center = 30
+            x_max = 32
+
+
+
+
+        print(self.device, self.is_half)
+
+        return x_pad, x_query, x_center, x_max
+
+# call inference
+class ClassVoices:
+    def __init__(self):
+        self.file_index = "" # root
+
+    def apply_conf(self, f0method,
+                   model_voice_path00, transpose00, file_index2_00,
+                   model_voice_path01, transpose01, file_index2_01,
+                   model_voice_path02, transpose02, file_index2_02,
+                   model_voice_path03, transpose03, file_index2_03,
+                   model_voice_path04, transpose04, file_index2_04,
+                   model_voice_path05, transpose05, file_index2_05,
+                   model_voice_path99, transpose99, file_index2_99):
+
+        #self.filename = filename
+        self.f0method = f0method # pm
+        
+        self.model_voice_path00 = model_voice_path00
+        self.transpose00 = transpose00
+        self.file_index200 = file_index2_00
+
+        self.model_voice_path01 = model_voice_path01
+        self.transpose01 = transpose01
+        self.file_index201 = file_index2_01
+
+        self.model_voice_path02 = model_voice_path02
+        self.transpose02 = transpose02
+        self.file_index202 = file_index2_02
+
+        self.model_voice_path03 = model_voice_path03
+        self.transpose03 = transpose03
+        self.file_index203 = file_index2_03
+
+        self.model_voice_path04 = model_voice_path04
+        self.transpose04 = transpose04
+        self.file_index204 = file_index2_04
+
+        self.model_voice_path05 = model_voice_path05
+        self.transpose05 = transpose05
+        self.file_index205 = file_index2_05
+
+        self.model_voice_path99 = model_voice_path99
+        self.transpose99 = transpose99
+        self.file_index299 = file_index2_99
+        return "CONFIGURATION APPLIED"
+
+    def custom_voice(self,
+        _values, # filter indices
+        audio_files, # all audio files
+        model_voice_path='',
+        transpose=0,
+        f0method='pm',
+        file_index='',
+        file_index2='',
+        ):
+
+        #hubert_model = None
+
+        get_vc(
+            sid=model_voice_path,  # model path
+            to_return_protect0=0.33,
+            to_return_protect1=0.33
+        )
+
+        for _value_item in _values:
+            filename = "audio2/"+audio_files[_value_item] if _value_item != "test" else audio_files[0]
+            #filename = "audio2/"+audio_files[_value_item]
+            try:
+                print(audio_files[_value_item], model_voice_path)
+            except:
+                pass
+
+            info_, (sample_, audio_output_) = vc_single(
+                sid=0,
+                input_audio_path=filename, #f"audio2/{filename}",
+                f0_up_key=transpose, # transpose for m to f and reverse 0 12
+                f0_file=None,
+                f0_method= f0method,
+                file_index= file_index, # dir pwd?
+                file_index2= file_index2,
+                # file_big_npy1,
+                index_rate= float(0.66),
+                filter_radius= int(3),
+                resample_sr= int(0),
+                rms_mix_rate= float(0.25),
+                protect= float(0.33),
+            )
+
+            sf.write(
+                file= filename, #f"audio2/{filename}",
+                samplerate=sample_,
+                data=audio_output_
+            )
+
+        # detele the model
+
+    def make_test(self, 
+        tts_text, 
+        tts_voice, 
+        model_path,
+        index_path,
+        transpose,
+        f0_method,
+        ):
+        os.system("rm -rf test")
+        filename = "test/test.wav"
+
+        if "SET_LIMIT" == os.getenv("DEMO"):
+          if len(tts_text) > 60:
+            tts_text = tts_text[:60]
+            print("DEMO; limit to 60 characters")
+
+        language = tts_voice[:2]
+        try:
+          os.system("mkdir test")
+          #nest_asyncio.apply() # gradio;not
+          asyncio.run(edge_tts.Communicate(tts_text, "-".join(tts_voice.split('-')[:-1])).save(filename))
+        except:
+          try:
+              tts = gTTS(tts_text, lang=language)
+              tts.save(filename)
+              tts.save
+              print(f'No audio was received. Please change the tts voice for {tts_voice}. USING gTTS.')
+          except:
+            tts = gTTS('a', lang=language)
+            tts.save(filename)
+            print('Error: Audio will be replaced.')
+
+        os.system("cp test/test.wav test/real_test.wav")
+
+        self([],[]) # start modules
+
+        self.custom_voice(
+            ["test"], # filter indices
+            ["test/test.wav"], # all audio files
+            model_voice_path=model_path,
+            transpose=transpose,
+            f0method=f0_method,
+            file_index='',
+            file_index2=index_path,
+        )
+        return "test/test.wav", "test/real_test.wav"
+
+    def __call__(self, speakers_list, audio_files):
+
+        speakers_indices = {}
+
+        for index, speak_ in enumerate(speakers_list):
+            if speak_ in speakers_indices:
+                speakers_indices[speak_].append(index)
+            else:
+                speakers_indices[speak_] = [index]
+
+        
+        # find models and index
+        global weight_root, index_root, config, hubert_model
+        weight_root = "weights"
+        names = []
+        for name in os.listdir(weight_root):
+            if name.endswith(".pth"):
+                names.append(name)
+
+        index_root = "logs"
+        index_paths = []
+        for name in os.listdir(index_root):
+            if name.endswith(".index"):
+                index_paths.append(name)
+
+        print(names, index_paths)
+        # config machine
+        hubert_model = None
+        config = Config('cuda:0', is_half=True) # config = Config('cpu', is_half=False) # cpu
+
+        # filter by speaker
+        for _speak, _values in speakers_indices.items():
+            #print(_speak, _values)
+            #for _value_item in _values:
+            #  self.filename = "audio2/"+audio_files[_value_item]
+            ###print(audio_files[_value_item])
+
+            #vc(_speak, _values, audio_files)
+
+            if _speak == "SPEAKER_00":
+              self.custom_voice(
+                    _values, # filteredd
+                    audio_files,
+                    model_voice_path=self.model_voice_path00,
+                    file_index2=self.file_index200,
+                    transpose=self.transpose00,
+                    f0method=self.f0method,
+                    file_index=self.file_index,
+                    )
+            elif _speak == "SPEAKER_01":
+                self.custom_voice(
+                    _values,
+                    audio_files,
+                    model_voice_path=self.model_voice_path01,
+                    file_index2=self.file_index201,
+                    transpose=self.transpose01,
+                    f0method=self.f0method,
+                    file_index=self.file_index,
+                )
+            elif _speak == "SPEAKER_02":
+                self.custom_voice(
+                    _values,
+                    audio_files,
+                    model_voice_path=self.model_voice_path02,
+                    file_index2=self.file_index202,
+                    transpose=self.transpose02,
+                    f0method=self.f0method,
+                    file_index=self.file_index,
+                )
+            elif _speak == "SPEAKER_03":
+                self.custom_voice(
+                    _values,
+                    audio_files,
+                    model_voice_path=self.model_voice_path03,
+                    file_index2=self.file_index203,
+                    transpose=self.transpose03,
+                    f0method=self.f0method,
+                    file_index=self.file_index,
+                )
+            elif _speak == "SPEAKER_04":
+                self.custom_voice(
+                    _values,
+                    audio_files,
+                    model_voice_path=self.model_voice_path04,
+                    file_index2=self.file_index204,
+                    transpose=self.transpose04,
+                    f0method=self.f0method,
+                    file_index=self.file_index,
+                )
+            elif _speak == "SPEAKER_05":
+                self.custom_voice(
+                    _values,
+                    audio_files,
+                    model_voice_path=self.model_voice_path05,
+                    file_index2=self.file_index205,
+                    transpose=self.transpose05,
+                    f0method=self.f0method,
+                    file_index=self.file_index,
+                )
+            elif _speak == "SPEAKER_99":
+                self.custom_voice(
+                    _values,
+                    audio_files,
+                    model_voice_path=self.model_voice_path99,
+                    file_index2=self.file_index299,
+                    transpose=self.transpose99,
+                    f0method=self.f0method,
+                    file_index=self.file_index,
+                )
+            else:
+                pass