Create rvc.py
Browse files- lib/rvc.py +451 -0
lib/rvc.py
ADDED
@@ -0,0 +1,451 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import argparse
|
2 |
+
import gc
|
3 |
+
import hashlib
|
4 |
+
import json
|
5 |
+
import os
|
6 |
+
import shlex
|
7 |
+
import subprocess
|
8 |
+
from contextlib import suppress
|
9 |
+
from urllib.parse import urlparse, parse_qs
|
10 |
+
|
11 |
+
import gradio as gr
|
12 |
+
import librosa
|
13 |
+
import numpy as np
|
14 |
+
import soundfile as sf
|
15 |
+
import sox
|
16 |
+
import yt_dlp
|
17 |
+
from pedalboard import Pedalboard, Reverb, Compressor, HighpassFilter
|
18 |
+
from pedalboard.io import AudioFile
|
19 |
+
from pydub import AudioSegment
|
20 |
+
from audio_separator.separator import Separator
|
21 |
+
from lib.infer import infer_audio
|
22 |
+
|
23 |
+
# Base directories
|
24 |
+
BASE_DIR = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
|
25 |
+
mdxnet_models_dir = os.path.join(BASE_DIR, 'mdxnet_models')
|
26 |
+
rvc_models_dir = os.path.join(BASE_DIR, 'models')
|
27 |
+
output_dir = os.path.join(BASE_DIR, 'song_output')
|
28 |
+
|
29 |
+
|
30 |
+
def get_youtube_video_id(url, ignore_playlist=True):
|
31 |
+
"""
|
32 |
+
Extract the YouTube video ID from various URL formats.
|
33 |
+
|
34 |
+
Examples:
|
35 |
+
http://youtu.be/SA2iWivDJiE
|
36 |
+
http://www.youtube.com/watch?v=_oPAwA_Udwc&feature=feedu
|
37 |
+
http://www.youtube.com/embed/SA2iWivDJiE
|
38 |
+
http://www.youtube.com/v/SA2iWivDJiE?version=3&hl=en_US
|
39 |
+
"""
|
40 |
+
parsed_url = urlparse(url)
|
41 |
+
hostname = parsed_url.hostname or ''
|
42 |
+
path = parsed_url.path
|
43 |
+
|
44 |
+
if hostname.lower() == 'youtu.be':
|
45 |
+
return path.lstrip('/')
|
46 |
+
|
47 |
+
if hostname.lower() in {'www.youtube.com', 'youtube.com', 'music.youtube.com'}:
|
48 |
+
if not ignore_playlist:
|
49 |
+
with suppress(KeyError):
|
50 |
+
return parse_qs(parsed_url.query)['list'][0]
|
51 |
+
if parsed_url.path == '/watch':
|
52 |
+
return parse_qs(parsed_url.query).get('v', [None])[0]
|
53 |
+
if parsed_url.path.startswith('/watch/'):
|
54 |
+
return parsed_url.path.split('/')[1]
|
55 |
+
if parsed_url.path.startswith('/embed/'):
|
56 |
+
return parsed_url.path.split('/')[2]
|
57 |
+
if parsed_url.path.startswith('/v/'):
|
58 |
+
return parsed_url.path.split('/')[2]
|
59 |
+
|
60 |
+
return None
|
61 |
+
|
62 |
+
|
63 |
+
def yt_download(link):
|
64 |
+
"""
|
65 |
+
Download the audio from a YouTube link as an mp3 file.
|
66 |
+
"""
|
67 |
+
ydl_opts = {
|
68 |
+
'format': 'bestaudio',
|
69 |
+
'outtmpl': '%(title)s',
|
70 |
+
'nocheckcertificate': True,
|
71 |
+
'ignoreerrors': True,
|
72 |
+
'no_warnings': True,
|
73 |
+
'quiet': True,
|
74 |
+
'extractaudio': True,
|
75 |
+
'postprocessors': [{
|
76 |
+
'key': 'FFmpegExtractAudio',
|
77 |
+
'preferredcodec': 'mp3'
|
78 |
+
}],
|
79 |
+
}
|
80 |
+
with yt_dlp.YoutubeDL(ydl_opts) as ydl:
|
81 |
+
result = ydl.extract_info(link, download=True)
|
82 |
+
download_path = ydl.prepare_filename(result, outtmpl='%(title)s.mp3')
|
83 |
+
return download_path
|
84 |
+
|
85 |
+
|
86 |
+
def display_progress(message, percent, is_webui, progress=None):
|
87 |
+
"""
|
88 |
+
Display progress either via the provided progress callback or by printing.
|
89 |
+
"""
|
90 |
+
if is_webui and progress is not None:
|
91 |
+
progress(percent, desc=message)
|
92 |
+
else:
|
93 |
+
print(message)
|
94 |
+
|
95 |
+
|
96 |
+
def raise_exception(error_msg, is_webui):
|
97 |
+
"""
|
98 |
+
Raise an exception. If running in a web UI, use gr.Error.
|
99 |
+
"""
|
100 |
+
if is_webui:
|
101 |
+
raise gr.Error(error_msg)
|
102 |
+
else:
|
103 |
+
raise Exception(error_msg)
|
104 |
+
|
105 |
+
|
106 |
+
|
107 |
+
def separation_uvr(filename, output):
|
108 |
+
"""
|
109 |
+
Run the separation steps using different pre-trained models.
|
110 |
+
Returns a tuple of four file paths:
|
111 |
+
- vocals_no_reverb: The vocals after initial de-echo/de-reverb (used as intermediate vocals)
|
112 |
+
- instrumental_path: The separated instrumental audio
|
113 |
+
- main_vocals_dereverb: The lead vocals after final de-reverb processing
|
114 |
+
- backup_vocals: The backup vocals extracted in the final stage
|
115 |
+
"""
|
116 |
+
separator = Separator(output_dir=output)
|
117 |
+
base_name = os.path.splitext(os.path.basename(filename))[0]
|
118 |
+
|
119 |
+
instrumental_path = os.path.join(output, f'{base_name}_Instrumental.wav')
|
120 |
+
initial_vocals = os.path.join(output, f'{base_name}_Vocals.wav')
|
121 |
+
vocals_no_reverb = os.path.join(output, f'{base_name}_Vocals (No Reverb).wav')
|
122 |
+
vocals_reverb = os.path.join(output, f'{base_name}_Vocals (Reverb).wav')
|
123 |
+
main_vocals_dereverb = os.path.join(output, f'{base_name}_Vocals_Main_DeReverb.wav')
|
124 |
+
backup_vocals = os.path.join(output, f'{base_name}_Vocals_Backup.wav')
|
125 |
+
|
126 |
+
separator.load_model(model_filename='model_bs_roformer_ep_317_sdr_12.9755.ckpt')
|
127 |
+
voc_inst = separator.separate(filename)
|
128 |
+
os.rename(os.path.join(output, voc_inst[0]), instrumental_path)
|
129 |
+
os.rename(os.path.join(output, voc_inst[1]), initial_vocals)
|
130 |
+
|
131 |
+
separator.load_model(model_filename='UVR-DeEcho-DeReverb.pth')
|
132 |
+
voc_no_reverb = separator.separate(initial_vocals)
|
133 |
+
os.rename(os.path.join(output, voc_no_reverb[0]), vocals_no_reverb)
|
134 |
+
os.rename(os.path.join(output, voc_no_reverb[1]), vocals_reverb)
|
135 |
+
|
136 |
+
separator.load_model(model_filename='mel_band_roformer_karaoke_aufr33_viperx_sdr_10.1956.ckpt')
|
137 |
+
voc_split = separator.separate(vocals_no_reverb)
|
138 |
+
os.rename(os.path.join(output, voc_split[0]), backup_vocals)
|
139 |
+
os.rename(os.path.join(output, voc_split[1]), main_vocals_dereverb)
|
140 |
+
|
141 |
+
if os.path.exists(vocals_reverb):
|
142 |
+
os.remove(vocals_reverb)
|
143 |
+
|
144 |
+
return vocals_no_reverb, instrumental_path, main_vocals_dereverb, backup_vocals
|
145 |
+
|
146 |
+
|
147 |
+
def get_audio_paths(song_dir):
|
148 |
+
"""
|
149 |
+
Search the given directory for expected audio files.
|
150 |
+
Returns:
|
151 |
+
orig_song_path, instrumentals_path, main_vocals_dereverb_path, backup_vocals_path
|
152 |
+
"""
|
153 |
+
orig_song_path = None
|
154 |
+
instrumentals_path = None
|
155 |
+
main_vocals_dereverb_path = None
|
156 |
+
backup_vocals_path = None
|
157 |
+
|
158 |
+
for file in os.listdir(song_dir):
|
159 |
+
if file.endswith('_Instrumental.wav'):
|
160 |
+
instrumentals_path = os.path.join(song_dir, file)
|
161 |
+
orig_song_path = instrumentals_path.replace('_Instrumental', '')
|
162 |
+
elif file.endswith('_Vocals_Main_DeReverb.wav'):
|
163 |
+
main_vocals_dereverb_path = os.path.join(song_dir, file)
|
164 |
+
elif file.endswith('_Vocals_Backup.wav'):
|
165 |
+
backup_vocals_path = os.path.join(song_dir, file)
|
166 |
+
|
167 |
+
return orig_song_path, instrumentals_path, main_vocals_dereverb_path, backup_vocals_path
|
168 |
+
|
169 |
+
|
170 |
+
def convert_to_stereo(audio_path):
|
171 |
+
"""
|
172 |
+
Convert the given audio file to stereo (2 channels) if it is mono.
|
173 |
+
"""
|
174 |
+
wave, sr = librosa.load(audio_path, mono=False, sr=44100)
|
175 |
+
if wave.ndim == 1:
|
176 |
+
stereo_path = f'{os.path.splitext(audio_path)[0]}_stereo.wav'
|
177 |
+
command = shlex.split(f'ffmpeg -y -loglevel error -i "{audio_path}" -ac 2 -f wav "{stereo_path}"')
|
178 |
+
subprocess.run(command, check=True)
|
179 |
+
return stereo_path
|
180 |
+
return audio_path
|
181 |
+
|
182 |
+
|
183 |
+
def pitch_shift(audio_path, pitch_change):
|
184 |
+
"""
|
185 |
+
Shift the pitch of the audio by the specified amount.
|
186 |
+
"""
|
187 |
+
output_path = f'{os.path.splitext(audio_path)[0]}_p{pitch_change}.wav'
|
188 |
+
if not os.path.exists(output_path):
|
189 |
+
y, sr = sf.read(audio_path)
|
190 |
+
tfm = sox.Transformer()
|
191 |
+
tfm.pitch(pitch_change)
|
192 |
+
y_shifted = tfm.build_array(input_array=y, sample_rate_in=sr)
|
193 |
+
sf.write(output_path, y_shifted, sr)
|
194 |
+
return output_path
|
195 |
+
|
196 |
+
|
197 |
+
def get_hash(filepath):
|
198 |
+
"""
|
199 |
+
Calculate a short BLAKE2b hash for the given file.
|
200 |
+
"""
|
201 |
+
with open(filepath, 'rb') as f:
|
202 |
+
file_hash = hashlib.blake2b()
|
203 |
+
while chunk := f.read(8192):
|
204 |
+
file_hash.update(chunk)
|
205 |
+
return file_hash.hexdigest()[:11]
|
206 |
+
|
207 |
+
|
208 |
+
def preprocess_song(song_input, song_id, is_webui, input_type, progress):
|
209 |
+
"""
|
210 |
+
Preprocess the input song:
|
211 |
+
- Download if YouTube URL.
|
212 |
+
- Convert to stereo.
|
213 |
+
- Separate vocals and instrumentals.
|
214 |
+
Returns a tuple with six values matching the expected unpacking in the pipeline.
|
215 |
+
"""
|
216 |
+
if input_type == 'yt':
|
217 |
+
display_progress('[~] Downloading song...', 0, is_webui, progress)
|
218 |
+
song_link = song_input.split('&')[0]
|
219 |
+
orig_song_path = yt_download(song_link)
|
220 |
+
elif input_type == 'local':
|
221 |
+
orig_song_path = song_input
|
222 |
+
else:
|
223 |
+
orig_song_path = None
|
224 |
+
|
225 |
+
song_output_dir = os.path.join(output_dir, song_id)
|
226 |
+
if not os.path.exists(song_output_dir):
|
227 |
+
os.makedirs(song_output_dir)
|
228 |
+
|
229 |
+
orig_song_path = convert_to_stereo(orig_song_path)
|
230 |
+
|
231 |
+
display_progress('[~] Separating Vocals from Instrumental...', 0.1, is_webui, progress)
|
232 |
+
vocals_no_reverb, instrumental_path, main_vocals_dereverb, backup_vocals = separation_uvr(orig_song_path, song_output_dir)
|
233 |
+
return orig_song_path, vocals_no_reverb, instrumental_path, main_vocals_dereverb, backup_vocals, main_vocals_dereverb
|
234 |
+
|
235 |
+
|
236 |
+
def voice_change(voice_model, vocals_path, output_path, pitch_change, f0_method,
|
237 |
+
index_rate, filter_radius, rms_mix_rate, protect, crepe_hop_length, is_webui):
|
238 |
+
"""
|
239 |
+
Convert the input vocals using the specified RVC model.
|
240 |
+
"""
|
241 |
+
rvc_model_path, rvc_index_path = get_rvc_model(voice_model, is_webui)
|
242 |
+
|
243 |
+
inferred_audio = infer_audio(
|
244 |
+
MODEL_NAME=voice_model,
|
245 |
+
SOUND_PATH=vocals_path,
|
246 |
+
F0_CHANGE=pitch_change,
|
247 |
+
F0_METHOD=f0_method,
|
248 |
+
CREPE_HOP_LENGTH=crepe_hop_length,
|
249 |
+
INDEX_RATE=index_rate,
|
250 |
+
FILTER_RADIUS=filter_radius,
|
251 |
+
RMS_MIX_RATE=rms_mix_rate,
|
252 |
+
PROTECT=protect,
|
253 |
+
)
|
254 |
+
gc.collect()
|
255 |
+
|
256 |
+
|
257 |
+
def add_audio_effects(audio_path, reverb_rm_size, reverb_wet, reverb_dry, reverb_damping):
|
258 |
+
"""
|
259 |
+
Apply a chain of audio effects (highpass, compression, reverb) to the input audio.
|
260 |
+
"""
|
261 |
+
output_path = f'{os.path.splitext(audio_path)[0]}_mixed.wav'
|
262 |
+
board = Pedalboard([
|
263 |
+
HighpassFilter(),
|
264 |
+
Compressor(ratio=4, threshold_db=-15),
|
265 |
+
Reverb(room_size=reverb_rm_size, dry_level=reverb_dry, wet_level=reverb_wet, damping=reverb_damping)
|
266 |
+
])
|
267 |
+
|
268 |
+
with AudioFile(audio_path) as f:
|
269 |
+
with AudioFile(output_path, 'w', f.samplerate, f.num_channels) as o:
|
270 |
+
while f.tell() < f.frames:
|
271 |
+
chunk = f.read(int(f.samplerate))
|
272 |
+
effected = board(chunk, f.samplerate, reset=False)
|
273 |
+
o.write(effected)
|
274 |
+
return output_path
|
275 |
+
|
276 |
+
|
277 |
+
def combine_audio(audio_paths, output_path, main_gain, backup_gain, inst_gain, output_format):
|
278 |
+
"""
|
279 |
+
Combine main vocals, backup vocals, and instrumental audio into a final mix.
|
280 |
+
"""
|
281 |
+
main_vocal_audio = AudioSegment.from_wav(audio_paths[0]) - 4 + main_gain
|
282 |
+
backup_vocal_audio = AudioSegment.from_wav(audio_paths[1]) - 6 + backup_gain
|
283 |
+
instrumental_audio = AudioSegment.from_wav(audio_paths[2]) - 7 + inst_gain
|
284 |
+
final_audio = main_vocal_audio.overlay(backup_vocal_audio).overlay(instrumental_audio)
|
285 |
+
final_audio.export(output_path, format=output_format)
|
286 |
+
|
287 |
+
|
288 |
+
def song_cover_pipeline(song_input, voice_model, pitch_change, keep_files,
|
289 |
+
is_webui=0, main_gain=0, backup_gain=0, inst_gain=0, index_rate=0.5, filter_radius=3,
|
290 |
+
rms_mix_rate=0.25, f0_method='rmvpe', crepe_hop_length=128, protect=0.33, pitch_change_all=0,
|
291 |
+
reverb_rm_size=0.15, reverb_wet=0.2, reverb_dry=0.8, reverb_damping=0.7, output_format='mp3',
|
292 |
+
progress=gr.Progress()):
|
293 |
+
"""
|
294 |
+
Main pipeline that orchestrates the AI cover song generation.
|
295 |
+
"""
|
296 |
+
try:
|
297 |
+
if not song_input or not voice_model:
|
298 |
+
raise_exception('Ensure that the song input field and voice model field is filled.', is_webui)
|
299 |
+
|
300 |
+
display_progress('[~] Starting AI Cover Generation Pipeline...', 0, is_webui, progress)
|
301 |
+
|
302 |
+
if urlparse(song_input).scheme == 'https':
|
303 |
+
input_type = 'yt'
|
304 |
+
song_id = get_youtube_video_id(song_input)
|
305 |
+
if song_id is None:
|
306 |
+
raise_exception('Invalid YouTube url.', is_webui)
|
307 |
+
else:
|
308 |
+
input_type = 'local'
|
309 |
+
song_input = song_input.strip('\"')
|
310 |
+
if os.path.exists(song_input):
|
311 |
+
song_id = get_hash(song_input)
|
312 |
+
else:
|
313 |
+
raise_exception(f'{song_input} does not exist.', is_webui)
|
314 |
+
|
315 |
+
song_dir = os.path.join(output_dir, song_id)
|
316 |
+
|
317 |
+
if not os.path.exists(song_dir):
|
318 |
+
os.makedirs(song_dir)
|
319 |
+
(orig_song_path, vocals_path, instrumentals_path,
|
320 |
+
main_vocals_path, backup_vocals_path, main_vocals_dereverb_path) = preprocess_song(
|
321 |
+
song_input, song_id, is_webui, input_type, progress
|
322 |
+
)
|
323 |
+
else:
|
324 |
+
vocals_path, main_vocals_path = None, None
|
325 |
+
paths = get_audio_paths(song_dir)
|
326 |
+
if any(path is None for path in paths) or keep_files:
|
327 |
+
(orig_song_path, vocals_path, instrumentals_path,
|
328 |
+
main_vocals_path, backup_vocals_path, main_vocals_dereverb_path) = preprocess_song(
|
329 |
+
song_input, song_id, is_webui, input_type, progress
|
330 |
+
)
|
331 |
+
else:
|
332 |
+
orig_song_path, instrumentals_path, main_vocals_dereverb_path, backup_vocals_path = paths
|
333 |
+
main_vocals_path = main_vocals_dereverb_path
|
334 |
+
|
335 |
+
pitch_change += pitch_change_all
|
336 |
+
|
337 |
+
base_song_name = os.path.splitext(os.path.basename(orig_song_path))[0]
|
338 |
+
algo_suffix = f"_{crepe_hop_length}" if f0_method == "mangio-crepe" else ""
|
339 |
+
ai_vocals_path = os.path.join(
|
340 |
+
song_dir,
|
341 |
+
f'{base_song_name}_lead_{voice_model}_p{pitch_change}_i{index_rate}_fr{filter_radius}_'
|
342 |
+
f'rms{rms_mix_rate}_pro{protect}_{f0_method}{algo_suffix}.wav'
|
343 |
+
)
|
344 |
+
ai_backing_path = os.path.join(
|
345 |
+
song_dir,
|
346 |
+
f'{base_song_name}_backing_{voice_model}_p{pitch_change}_i{index_rate}_fr{filter_radius}_'
|
347 |
+
f'rms{rms_mix_rate}_pro{protect}_{f0_method}{algo_suffix}.wav'
|
348 |
+
)
|
349 |
+
ai_cover_path = os.path.join(song_dir, f'{base_song_name} ({voice_model} Ver).{output_format}')
|
350 |
+
ai_cover_backing_path = os.path.join(song_dir, f'{base_song_name} ({voice_model} Ver With Backing).{output_format}')
|
351 |
+
|
352 |
+
if not os.path.exists(ai_vocals_path):
|
353 |
+
display_progress('[~] Converting lead voice using RVC...', 0.5, is_webui, progress)
|
354 |
+
voice_change(voice_model, main_vocals_dereverb_path, ai_vocals_path, pitch_change,
|
355 |
+
f0_method, index_rate, filter_radius, rms_mix_rate, protect, crepe_hop_length, is_webui)
|
356 |
+
|
357 |
+
display_progress('[~] Converting backing voice using RVC...', 0.65, is_webui, progress)
|
358 |
+
voice_change(voice_model, backup_vocals_path, ai_backing_path, pitch_change,
|
359 |
+
f0_method, index_rate, filter_radius, rms_mix_rate, protect, crepe_hop_length, is_webui)
|
360 |
+
|
361 |
+
display_progress('[~] Applying audio effects to Vocals...', 0.8, is_webui, progress)
|
362 |
+
ai_vocals_mixed_path = add_audio_effects(ai_vocals_path, reverb_rm_size, reverb_wet, reverb_dry, reverb_damping)
|
363 |
+
ai_backing_mixed_path = add_audio_effects(ai_backing_path, reverb_rm_size, reverb_wet, reverb_dry, reverb_damping)
|
364 |
+
|
365 |
+
if pitch_change_all != 0:
|
366 |
+
display_progress('[~] Applying overall pitch change', 0.85, is_webui, progress)
|
367 |
+
instrumentals_path = pitch_shift(instrumentals_path, pitch_change_all)
|
368 |
+
backup_vocals_path = pitch_shift(backup_vocals_path, pitch_change_all)
|
369 |
+
|
370 |
+
display_progress('[~] Combining AI Vocals and Instrumentals...', 0.9, is_webui, progress)
|
371 |
+
combine_audio([ai_vocals_mixed_path, backup_vocals_path, instrumentals_path],
|
372 |
+
ai_cover_path, main_gain, backup_gain, inst_gain, output_format)
|
373 |
+
combine_audio([ai_vocals_mixed_path, ai_backing_mixed_path, instrumentals_path],
|
374 |
+
ai_cover_backing_path, main_gain, backup_gain, inst_gain, output_format)
|
375 |
+
|
376 |
+
if not keep_files:
|
377 |
+
display_progress('[~] Removing intermediate audio files...', 0.95, is_webui, progress)
|
378 |
+
intermediate_files = [vocals_path, main_vocals_path, ai_vocals_mixed_path, ai_backing_mixed_path]
|
379 |
+
if pitch_change_all != 0:
|
380 |
+
intermediate_files += [instrumentals_path, backup_vocals_path]
|
381 |
+
for file in intermediate_files:
|
382 |
+
if file and os.path.exists(file):
|
383 |
+
os.remove(file)
|
384 |
+
|
385 |
+
return ai_cover_path, ai_cover_backing_path
|
386 |
+
|
387 |
+
except Exception as e:
|
388 |
+
raise_exception(str(e), is_webui)
|
389 |
+
|
390 |
+
|
391 |
+
if __name__ == '__main__':
|
392 |
+
parser = argparse.ArgumentParser(
|
393 |
+
description='AICoverGen: Mod.',
|
394 |
+
add_help=True
|
395 |
+
)
|
396 |
+
parser.add_argument('-i', '--song-input', type=str, required=True,
|
397 |
+
help='Link to a YouTube video or the filepath to a local mp3/wav file to create an AI cover of')
|
398 |
+
parser.add_argument('-dir', '--rvc-dirname', type=str, required=True,
|
399 |
+
help='Name of the folder in the rvc_models directory containing the RVC model file and optional index file to use')
|
400 |
+
parser.add_argument('-p', '--pitch-change', type=int, required=True,
|
401 |
+
help='Change the pitch of AI Vocals only. Generally, use 1 for male to female and -1 for vice-versa. (Octaves)')
|
402 |
+
parser.add_argument('-k', '--keep-files', action=argparse.BooleanOptionalAction,
|
403 |
+
help='Whether to keep all intermediate audio files generated in the song_output/id directory, e.g. Isolated Vocals/Instrumentals')
|
404 |
+
parser.add_argument('-ir', '--index-rate', type=float, default=0.5,
|
405 |
+
help='A decimal number e.g. 0.5, used to reduce/resolve the timbre leakage problem. If set to 1, more biased towards the timbre quality of the training dataset')
|
406 |
+
parser.add_argument('-fr', '--filter-radius', type=int, default=3,
|
407 |
+
help='A number between 0 and 7. If >=3: apply median filtering to the harvested pitch results. The value represents the filter radius and can reduce breathiness.')
|
408 |
+
parser.add_argument('-rms', '--rms-mix-rate', type=float, default=0.25,
|
409 |
+
help="A decimal number e.g. 0.25. Control how much to use the original vocal's loudness (0) or a fixed loudness (1).")
|
410 |
+
parser.add_argument('-palgo', '--pitch-detection-algo', type=str, default='rmvpe',
|
411 |
+
help='Best option is rmvpe (clarity in vocals), then mangio-crepe (smoother vocals).')
|
412 |
+
parser.add_argument('-hop', '--crepe-hop-length', type=int, default=128,
|
413 |
+
help='If pitch detection algo is mangio-crepe, controls how often it checks for pitch changes in milliseconds. Recommended: 128.')
|
414 |
+
parser.add_argument('-pro', '--protect', type=float, default=0.33,
|
415 |
+
help='A decimal number e.g. 0.33. Protect voiceless consonants and breath sounds to prevent artifacts such as tearing in electronic music.')
|
416 |
+
parser.add_argument('-mv', '--main-vol', type=int, default=0,
|
417 |
+
help='Volume change for AI main vocals in decibels. Use -3 to decrease by 3 dB and 3 to increase by 3 dB')
|
418 |
+
parser.add_argument('-bv', '--backup-vol', type=int, default=0,
|
419 |
+
help='Volume change for backup vocals in decibels')
|
420 |
+
parser.add_argument('-iv', '--inst-vol', type=int, default=0,
|
421 |
+
help='Volume change for instrumentals in decibels')
|
422 |
+
parser.add_argument('-pall', '--pitch-change-all', type=int, default=0,
|
423 |
+
help='Change the pitch/key of vocals and instrumentals. Changing this slightly reduces sound quality')
|
424 |
+
parser.add_argument('-rsize', '--reverb-size', type=float, default=0.15,
|
425 |
+
help='Reverb room size between 0 and 1')
|
426 |
+
parser.add_argument('-rwet', '--reverb-wetness', type=float, default=0.2,
|
427 |
+
help='Reverb wet level between 0 and 1')
|
428 |
+
parser.add_argument('-rdry', '--reverb-dryness', type=float, default=0.8,
|
429 |
+
help='Reverb dry level between 0 and 1')
|
430 |
+
parser.add_argument('-rdamp', '--reverb-damping', type=float, default=0.7,
|
431 |
+
help='Reverb damping between 0 and 1')
|
432 |
+
parser.add_argument('-oformat', '--output-format', type=str, default='mp3',
|
433 |
+
help='Output format of audio file. mp3 for smaller file size, wav for best quality')
|
434 |
+
args = parser.parse_args()
|
435 |
+
|
436 |
+
rvc_dir = os.path.join(rvc_models_dir, args.rvc_dirname)
|
437 |
+
if not os.path.exists(rvc_dir):
|
438 |
+
raise Exception(f'The folder {rvc_dir} does not exist.')
|
439 |
+
|
440 |
+
cover_path, cover_with_backing = song_cover_pipeline(
|
441 |
+
args.song_input, args.rvc_dirname, args.pitch_change, args.keep_files,
|
442 |
+
main_gain=args.main_vol, backup_gain=args.backup_vol, inst_gain=args.inst_vol,
|
443 |
+
index_rate=args.index_rate, filter_radius=args.filter_radius,
|
444 |
+
rms_mix_rate=args.rms_mix_rate, f0_method=args.pitch_detection_algo,
|
445 |
+
crepe_hop_length=args.crepe_hop_length, protect=args.protect,
|
446 |
+
pitch_change_all=args.pitch_change_all,
|
447 |
+
reverb_rm_size=args.reverb_size, reverb_wet=args.reverb_wetness,
|
448 |
+
reverb_dry=args.reverb_dryness, reverb_damping=args.reverb_damping,
|
449 |
+
output_format=args.output_format
|
450 |
+
)
|
451 |
+
print(f'[+] Cover generated at {cover_path}')
|