File size: 15,495 Bytes
d6895d6
 
3f22226
 
d6895d6
 
 
 
 
 
 
 
7920a69
 
2572108
d6895d6
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
7920a69
d6895d6
1209356
 
 
 
 
 
 
d6895d6
7920a69
 
d6895d6
 
 
 
 
 
 
 
 
eacae25
d6895d6
 
 
 
afaa295
 
 
5db6bc1
 
 
 
d6895d6
 
 
 
7920a69
 
 
 
5db6bc1
7920a69
5db6bc1
7920a69
3f22226
 
 
 
d6895d6
 
3f22226
7920a69
d6895d6
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
9434313
d6895d6
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
9434313
d6895d6
 
 
 
 
 
 
 
 
 
 
9434313
d6895d6
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
9434313
d6895d6
 
 
 
9434313
d6895d6
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
9434313
d6895d6
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
9434313
d6895d6
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
9434313
d6895d6
 
 
 
9434313
d6895d6
 
 
 
 
9434313
d6895d6
 
9434313
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
import sys
import io, os, stat
import torch

import subprocess
import random
from zipfile import ZipFile
import uuid
import time
import torchaudio
import numpy as np

# update gradio to faster streaming
# download for mecab
print("install unidic")
os.system('python -m unidic download')

# By using XTTS you agree to CPML license https://coqui.ai/cpml
os.environ["COQUI_TOS_AGREED"] = "1"

# langid is used to detect language for longer text
# Most users expect text to be their own language, there is checkbox to disable it
import langid
import base64
import csv
from io import StringIO
import datetime
import re

from scipy.io.wavfile import write
from pydub import AudioSegment

from TTS.api import TTS
from TTS.tts.configs.xtts_config import XttsConfig
from TTS.tts.models.xtts import Xtts
from TTS.utils.generic_utils import get_user_data_dir
from huggingface_hub import HfApi

# Use never ffmpeg binary for Ubuntu20 to use denoising for microphone input
print("Export newer ffmpeg binary for denoise filter")
ZipFile("ffmpeg.zip").extractall()
print("Make ffmpeg binary executable")
st = os.stat("ffmpeg")
os.chmod("ffmpeg", st.st_mode | stat.S_IEXEC)

HF_TOKEN = os.environ.get("HF_TOKEN")
if not HF_TOKEN:
    raise ValueError("HF_TOKEN environment variable is not set")

# will use api to restart space on a unrecoverable error
api = HfApi(token=HF_TOKEN)
repo_id = "coqui/xtts"

# This will trigger downloading model
print("Downloading if not downloaded Coqui XTTS V2")
from TTS.utils.manage import ModelManager

model_name = "tts_models/multilingual/multi-dataset/xtts_v2"
ModelManager().download_model(model_name)
model_path = os.path.join(get_user_data_dir("tts"), model_name.replace("/", "--"))
print("XTTS downloaded")

# Ensure the model path and its contents are accessible
os.system(f'chown -R appuser:appgroup {model_path}')
os.system(f'chmod -R 755 {model_path}')
# Ensure the model directory and files have the correct permissions
if not os.access(model_path, os.W_OK):
    raise PermissionError(f"Write permission denied for model directory: {model_path}")

config = XttsConfig()
config.load_json(os.path.join(model_path, "config.json"))

model = Xtts.init_from_config(config)
checkpoint_path = os.path.join(model_path, "model.pth")
vocab_path = os.path.join(model_path, "vocab.json")

if not os.path.exists(checkpoint_path):
    raise FileNotFoundError(f"Checkpoint file not found at {checkpoint_path}")
if not os.path.exists(vocab_path):
    raise FileNotFoundError(f"Vocab file not found at {vocab_path}")

if not os.environ.get('CUDA_HOME'):
    print(f"ENV var CUDA_HOME is not set, defaulting to: '/usr/local/cuda'")
    os.environ['CUDA_HOME'] = f"/usr/local/cuda"

model.load_checkpoint(
    config,
    checkpoint_dir=model_path,
    vocab_path=vocab_path,
    eval=True,
    use_deepspeed=True,
)
model.cuda()

# This is for debugging purposes only
DEVICE_ASSERT_DETECTED = 0
DEVICE_ASSERT_PROMPT = None
DEVICE_ASSERT_LANG = None

supported_languages = config.languages
def numpy_to_mp3(audio_array, sampling_rate):
    # Normalize audio_array if it's floating-point
    if np.issubdtype(audio_array.dtype, np.floating):
        max_val = np.max(np.abs(audio_array))
        audio_array = (audio_array / max_val) * 32767  # Normalize to 16-bit range
        audio_array = audio_array.astype(np.int16)

    # Create an audio segment from the numpy array
    audio_segment = AudioSegment(
        audio_array.tobytes(),
        frame_rate=sampling_rate,
        sample_width=audio_array.dtype.itemsize,
        channels=1
    )

    # Export the audio segment to MP3 bytes - use a high bitrate to maximise quality
    mp3_io = io.BytesIO()
    audio_segment.export(mp3_io, format="mp3", bitrate="320k")

    # Get the MP3 bytes
    mp3_bytes = mp3_io.getvalue()
    mp3_io.close()

    return mp3_bytes

def predict(
    prompt,
    language,
    audio_file_pth,
    mic_file_path,
    use_mic,
    voice_cleanup,
    no_lang_auto_detect,
    agree,
):
    if agree == True:
        if language not in supported_languages:
            print(
                f"Language you put {language} in is not in is not in our Supported Languages, please choose from dropdown"
            )

            return (
                None,
            )

        language_predicted = langid.classify(prompt)[
            0
        ].strip()  # strip need as there is space at end!

        # tts expects chinese as zh-cn
        if language_predicted == "zh":
            # we use zh-cn
            language_predicted = "zh-cn"

        print(f"Detected language:{language_predicted}, Chosen language:{language}")

        # After text character length 15 trigger language detection
        if len(prompt) > 15:
            # allow any language for short text as some may be common
            # If user unchecks language autodetection it will not trigger
            # You may remove this completely for own use
            if language_predicted != language and not no_lang_auto_detect:
                # Please duplicate and remove this check if you really want this
                # Or auto-detector fails to identify language (which it can on pretty short text or mixed text)
                print(
                    f"It looks like your text isn’t the language you chose , if you’re sure the text is the same language you chose, please check disable language auto-detection checkbox"
                )

                return (
                    None,
                )

        if use_mic == True:
            if mic_file_path is not None:
                speaker_wav = mic_file_path
            else:
                print(
                    "Please record your voice with Microphone, or uncheck Use Microphone to use reference audios"
                )
                return (
                    None,
                )

        else:
            speaker_wav = audio_file_pth

        # Filtering for microphone input, as it has BG noise, maybe silence in beginning and end
        # This is fast filtering not perfect

        # Apply all on demand
        lowpassfilter = denoise = trim = loudness = True

        if lowpassfilter:
            lowpass_highpass = "lowpass=8000,highpass=75,"
        else:
            lowpass_highpass = ""

        if trim:
            # better to remove silence in beginning and end for microphone
            trim_silence = "areverse,silenceremove=start_periods=1:start_silence=0:start_threshold=0.02,areverse,silenceremove=start_periods=1:start_silence=0:start_threshold=0.02,"
        else:
            trim_silence = ""

        if voice_cleanup:
            try:
                out_filename = (
                    speaker_wav + str(uuid.uuid4()) + ".wav"
                )  # ffmpeg to know output format

                # we will use newer ffmpeg as that has afftn denoise filter
                shell_command = f"./ffmpeg -y -i {speaker_wav} -af {lowpass_highpass}{trim_silence} {out_filename}".split(
                    " "
                )

                command_result = subprocess.run(
                    [item for item in shell_command],
                    capture_output=False,
                    text=True,
                    check=True,
                )
                speaker_wav = out_filename
                print("Filtered microphone input")
            except subprocess.CalledProcessError:
                # There was an error - command exited with non-zero code
                print("Error: failed filtering, use original microphone input")
        else:
            speaker_wav = speaker_wav

        if len(prompt) < 2:
            print("Please give a longer prompt text")
            return (
                None,
            )
        if len(prompt) > 1000:
            print(
                "Text length limited to 200 characters for this demo, please try shorter text. You can clone this space and edit code for your own usage"
            )
            return (
                None,
            )
        global DEVICE_ASSERT_DETECTED
        if DEVICE_ASSERT_DETECTED:
            global DEVICE_ASSERT_PROMPT
            global DEVICE_ASSERT_LANG
            # It will likely never come here as we restart space on first unrecoverable error now
            print(
                f"Unrecoverable exception caused by language:{DEVICE_ASSERT_LANG} prompt:{DEVICE_ASSERT_PROMPT}"
            )

            # HF Space specific.. This error is unrecoverable need to restart space
            space = api.get_space_runtime(repo_id=repo_id)
            if space.stage != "BUILDING":
                api.restart_space(repo_id=repo_id)
            else:
                print("TRIED TO RESTART but space is building")

        try:
            metrics_text = ""
            t_latent = time.time()

            # note diffusion_conditioning not used on hifigan (default mode), it will be empty but need to pass it to model.inference
            try:
                (
                    gpt_cond_latent,
                    speaker_embedding,
                ) = model.get_conditioning_latents(audio_path=speaker_wav, gpt_cond_len=30, gpt_cond_chunk_len=4, max_ref_length=60)
            except Exception as e:
                print("Speaker encoding error", str(e))
                print(
                    "It appears something wrong with reference, did you unmute your microphone?"
                )
                return (
                    None,
                )

            latent_calculation_time = time.time() - t_latent
            # metrics_text=f"Embedding calculation time: {latent_calculation_time:.2f} seconds\n"

            # temporary comma fix
            prompt = re.sub("([^\x00-\x7F]|\w)(\.|\。|\?)", r"\1 \2\2", prompt)

            wav_chunks = []
            ## Direct mode
            """
            print("I: Generating new audio...")
            t0 = time.time()
            out = model.inference(
                prompt,
                language,
                gpt_cond_latent,
                speaker_embedding,
                repetition_penalty=5.0,
                temperature=0.75,
            )
            inference_time = time.time() - t0
            print(f"I: Time to generate audio: {round(inference_time*1000)} milliseconds")
            metrics_text+=f"Time to generate audio: {round(inference_time*1000)} milliseconds\n"
            real_time_factor= (time.time() - t0) / out['wav'].shape[-1] * 24000
            print(f"Real-time factor (RTF): {real_time_factor}")
            metrics_text+=f"Real-time factor (RTF): {real_time_factor:.2f}\n"
            torchaudio.save("output.wav", torch.tensor(out["wav"]).unsqueeze(0), 24000)
            """
            print("I: Generating new audio in streaming mode...")
            t0 = time.time()
            chunks = model.inference_stream(
                prompt,
                language,
                gpt_cond_latent,
                speaker_embedding,
                repetition_penalty=7.0,
                temperature=0.85,
            )

            first_chunk = True
            for i, chunk in enumerate(chunks):
                if first_chunk:
                    first_chunk_time = time.time() - t0
                    metrics_text += f"Latency to first audio chunk: {round(first_chunk_time*1000)} milliseconds\n"
                    first_chunk = False

                # Convert chunk to numpy array and return it
                chunk_np = chunk.cpu().numpy()
                print('chunk',i)
                yield (24000, chunk_np)
                wav_chunks.append(chunk)
                
                print(f"Received chunk {i} of audio length {chunk.shape[-1]}")
            inference_time = time.time() - t0
            print(
                f"I: Time to generate audio: {round(inference_time*1000)} milliseconds"
            )
            # metrics_text += (
            #    f"Time to generate audio: {round(inference_time*1000)} milliseconds\n"
            #)

        except RuntimeError as e:
            if "device-side assert" in str(e):
                # cannot do anything on cuda device side error, need tor estart
                print(
                    f"Exit due to: Unrecoverable exception caused by language:{language} prompt:{prompt}",
                    flush=True,
                )
                print("Unhandled Exception encounter, please retry in a minute")
                print("Cuda device-assert Runtime encountered need restart")
                if not DEVICE_ASSERT_DETECTED:
                    DEVICE_ASSERT_DETECTED = 1
                    DEVICE_ASSERT_PROMPT = prompt
                    DEVICE_ASSERT_LANG = language

                # just before restarting save what caused the issue so we can handle it in future
                # Uploading Error data only happens for unrecovarable error
                error_time = datetime.datetime.now().strftime("%d-%m-%Y-%H:%M:%S")
                error_data = [
                    error_time,
                    prompt,
                    language,
                    audio_file_pth,
                    mic_file_path,
                    use_mic,
                    voice_cleanup,
                    no_lang_auto_detect,
                    agree,
                ]
                error_data = [str(e) if type(e) != str else e for e in error_data]
                print(error_data)
                print(speaker_wav)
                write_io = StringIO()
                csv.writer(write_io).writerows([error_data])
                csv_upload = write_io.getvalue().encode()

                filename = error_time + "_" + str(uuid.uuid4()) + ".csv"
                print("Writing error csv")
                error_api = HfApi()
                error_api.upload_file(
                    path_or_fileobj=csv_upload,
                    path_in_repo=filename,
                    repo_id="coqui/xtts-flagged-dataset",
                    repo_type="dataset",
                )

                # speaker_wav
                print("Writing error reference audio")
                speaker_filename = (
                    error_time + "_reference_" + str(uuid.uuid4()) + ".wav"
                )
                error_api = HfApi()
                error_api.upload_file(
                    path_or_fileobj=speaker_wav,
                    path_in_repo=speaker_filename,
                    repo_id="coqui/xtts-flagged-dataset",
                    repo_type="dataset",
                )

                # HF Space specific.. This error is unrecoverable need to restart space
                space = api.get_space_runtime(repo_id=repo_id)
                if space.stage != "BUILDING":
                    api.restart_space(repo_id=repo_id)
                else:
                    print("TRIED TO RESTART but space is building")

            else:
                if "Failed to decode" in str(e):
                    print("Speaker encoding error", str(e))
                    print(
                        "It appears something wrong with reference, did you unmute your microphone?"
                    )
                else:
                    print("RuntimeError: non device-side assert error:", str(e))
                    print("Something unexpected happened please retry again.")
                return (
                    None,
                )

    else:
        print("Please accept the Terms & Condition!")
        return (
            None,
        )