File size: 15,689 Bytes
ca90f09
f81d4f2
0db6209
f05e79d
0db6209
5b0db06
f81d4f2
 
 
2f7d9da
 
d02ad9c
0db6209
 
f89c55d
 
d02ad9c
f81d4f2
 
 
194fffd
f81d4f2
 
 
 
def995e
f81d4f2
def995e
f81d4f2
def995e
 
 
 
0db6209
 
 
 
 
 
 
 
0d1adfa
194fffd
f81d4f2
 
 
 
 
 
 
 
 
 
 
 
0db6209
 
def995e
 
 
 
3b69bc5
ee48acc
f89c55d
f81d4f2
f89c55d
f81d4f2
f89c55d
 
 
 
3b69bc5
f81d4f2
f89c55d
 
 
 
0db6209
f89c55d
 
 
63c2202
 
0db6209
f38c6b2
0db6209
 
 
f38c6b2
63c2202
 
f38c6b2
f89c55d
63c2202
 
 
3b69bc5
f81d4f2
63c2202
f89c55d
 
f74bce2
 
3b69bc5
f74bce2
 
 
 
 
3b69bc5
f81d4f2
f74bce2
 
 
 
3b69bc5
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
0db6209
3b69bc5
 
 
 
 
 
 
 
 
 
 
 
 
 
 
f74bce2
 
 
 
 
 
3b69bc5
f81d4f2
f74bce2
2fb935b
f38c6b2
2fb935b
 
 
3b69bc5
f81d4f2
2fb935b
def995e
 
 
 
f38c6b2
def995e
 
f74bce2
f81d4f2
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
f74bce2
f81d4f2
 
 
 
 
 
 
 
d6aee2b
def995e
d6aee2b
def995e
d6aee2b
 
def995e
 
 
 
 
 
 
 
d6aee2b
def995e
d6aee2b
939c1fe
 
 
 
 
f81d4f2
3b69bc5
d636635
ee48acc
 
f74bce2
 
 
3b69bc5
f81d4f2
f74bce2
194fffd
 
fc50d18
194fffd
e1c65f1
12438dd
533ef97
e886026
533ef97
e886026
533ef97
e886026
533ef97
e1c65f1
 
 
 
 
f89c55d
 
 
 
 
f81d4f2
e1c65f1
 
3444a7f
 
 
 
 
0dfedcd
d636635
869b784
d636635
 
f74bce2
 
f38c6b2
3b69bc5
939c1fe
3b69bc5
d636635
0dfedcd
f74bce2
0dfedcd
 
f74bce2
 
f38c6b2
3b69bc5
939c1fe
0dfedcd
e886026
 
 
 
 
 
f38c6b2
3b69bc5
e886026
 
 
 
 
 
 
 
f38c6b2
3b69bc5
e886026
 
 
 
 
 
 
 
f38c6b2
3b69bc5
e886026
 
 
 
 
 
 
 
f38c6b2
3b69bc5
e886026
 
d636635
f74bce2
d636635
 
f74bce2
3b69bc5
f74bce2
f38c6b2
f74bce2
 
 
 
 
 
 
 
f38c6b2
3b69bc5
939c1fe
d636635
e886026
 
 
 
 
3b69bc5
e886026
f38c6b2
e886026
 
 
 
 
 
 
3b69bc5
e886026
f38c6b2
e886026
 
 
 
 
 
 
3b69bc5
e886026
f38c6b2
e886026
 
 
 
 
 
 
3b69bc5
e886026
f38c6b2
e886026
 
0dfedcd
 
e886026
 
194fffd
 
 
142fdc7
 
33aef52
3c140e1
142fdc7
 
 
 
 
 
 
 
 
 
 
 
 
 
 
e886026
142fdc7
c6f8b29
142fdc7
 
0dfedcd
142fdc7
 
 
262bc3f
142fdc7
939c1fe
 
f74bce2
 
 
 
3b69bc5
0287131
 
3b69bc5
 
 
 
f38c6b2
 
c819034
939c1fe
 
58c1f3f
939c1fe
142fdc7
3b69bc5
 
194fffd
 
98d97a5
f81d4f2
 
3b69bc5
194fffd
 
cff1c92
3444a7f
0f6c489
8fbd19a
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
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
import sys
import io, os, stat
import subprocess
import random
from zipfile import ZipFile
import uuid 
import time
import torch
import torchaudio
# 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 gradio as gr
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

HF_TOKEN = os.environ.get("HF_TOKEN")

from huggingface_hub import HfApi

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

# 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)

# Load TTS
tts = TTS("tts_models/multilingual/multi-dataset/xtts_v1")

model_path = os.path.join(get_user_data_dir("tts"), "tts_models--multilingual--multi-dataset--xtts_v1")
config = XttsConfig()
config.load_json(os.path.join(model_path, "config.json"))
model = Xtts.init_from_config(config)
model.load_checkpoint(
    config,
    checkpoint_path=os.path.join(model_path, "model.pth"),
    vocab_path=os.path.join(model_path, "vocab.json"),
    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

def predict(prompt, language, audio_file_pth, mic_file_path, use_mic, voice_cleanup, no_lang_auto_detect, agree,):
    if agree == True:
        supported_languages=["en","es","fr","de","it","pt","pl","tr","ru","nl","cs","ar","zh-cn"]
        
        if language not in supported_languages:
            gr.Warning(f"Language you put {language} in is not in is not in our Supported Languages, please choose from dropdown")
                
            return (
                    None,
                    None,
                    None,
                    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)
                gr.Warning(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,
                        None,
                        None,
                        None,
                    ) 

        
        if use_mic == True:
            if mic_file_path is not None:
               speaker_wav=mic_file_path
            else:
                gr.Warning("Please record your voice with Microphone, or uncheck Use Microphone to use reference audios")
                return (
                    None,
                    None,
                    None,
                    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:
            gr.Warning("Please give a longer prompt text")
            return (
                    None,
                    None,
                    None,
                    None,
                )
        if len(prompt)>200:
            gr.Warning("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,
                    None,
                    None,
                    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}")
            
        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
            gpt_cond_latent, diffusion_conditioning, speaker_embedding = model.get_conditioning_latents(audio_path=speaker_wav)
            latent_calculation_time = time.time() - t_latent
            #metrics_text=f"Embedding calculation time: {latent_calculation_time:.2f} seconds\n"
            
            wav_chunks = []
    
            print("I: Generating new audio...")
            t0 = time.time()
            out = model.inference(
                prompt,
                language,
                gpt_cond_latent,
                speaker_embedding,
                diffusion_conditioning
            )
            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)
            
        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)
                gr.Warning("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

                
                # HF Space specific.. This error is unrecoverable need to restart space 
                api.restart_space(repo_id=repo_id)
            else:
                print("RuntimeError: non device-side assert error:", str(e))
                raise e
        return (
            gr.make_waveform(
                audio="output.wav",
            ),
            "output.wav",
            metrics_text,
            speaker_wav,
        )
    else:
        gr.Warning("Please accept the Terms & Condition!")
        return (
                None,
                None,
                None,
                None,
            ) 


title = "Coqui🐸 XTTS"

description = """
<a href="https://huggingface.co/coqui/XTTS-v1">XTTS</a> is a Voice generation model that lets you clone voices into different languages by using just a quick 6-second audio clip. 
<br/>
XTTS is built on previous research, like Tortoise, with additional architectural innovations and training to make cross-language voice cloning and multilingual speech generation possible. 
<br/>
This is the same model that powers our creator application <a href="https://coqui.ai">Coqui Studio</a> as well as the <a href="https://docs.coqui.ai">Coqui API</a>. In production we apply modifications to make low-latency streaming possible.
<br/>
Leave a star on the Github <a href="https://github.com/coqui-ai/TTS">🐸TTS</a>, where our open-source inference and training code lives.
<br/>
<p>For faster inference without waiting in the queue, you should duplicate this space and upgrade to GPU via the settings.
<br/>
<a href="https://huggingface.co/spaces/coqui/xtts?duplicate=true">
<img style="margin-top: 0em; margin-bottom: 0em" src="https://bit.ly/3gLdBN6" alt="Duplicate Space"></a>
</p>
<p>Language Selectors: 
Arabic: ar, Brazilian Portuguese: pt , Chinese: zh-cn, Czech: cs,<br/> 
Dutch: nl, English: en, French: fr, Italian: it, Polish: pl,<br/> 
Russian: ru, Spanish: es, Turkish: tr <br/> 
</p>
<img referrerpolicy="no-referrer-when-downgrade" src="https://static.scarf.sh/a.png?x-pxid=8946ef36-c454-4a8e-a9c9-8a8dd735fabd" />
"""

article = """
<div style='margin:20px auto;'>
<p>By using this demo you agree to the terms of the Coqui Public Model License at https://coqui.ai/cpml</p>
</div>
"""
examples = [
    [
        "Once when I was six years old I saw a magnificent picture",
        "en",
        "examples/female.wav",
        None,
        False,
        False,
        False,
        True,

    ],
    [
        "Lorsque j'avais six ans j'ai vu, une fois, une magnifique image",
        "fr",
        "examples/male.wav",
        None,
        False,
        False,
        False,
        True,
    ],
    [
        "Als ich sechs war, sah ich einmal ein wunderbares Bild",
        "de",
        "examples/female.wav",
        None,
        False,
        False,
        False,
        True,
    ],
    [
        "Cuando tenía seis años, vi una vez una imagen magnífica",
        "es",
        "examples/male.wav",
        None,
        False,
        False,
        False,
        True,
    ],
    [
        "Quando eu tinha seis anos eu vi, uma vez, uma imagem magnífica",
        "pt",
        "examples/female.wav",
        None,
        False,
        False,
        False,
        True,
    ],
    [
        "Kiedy miałem sześć lat, zobaczyłem pewnego razu wspaniały obrazek",
        "pl",
        "examples/male.wav",
        None,
        False,
        False,
        False,
        True,
    ],
    [
        "Un tempo lontano, quando avevo sei anni, vidi un magnifico disegno",
        "it",
        "examples/female.wav",
        None,
       False,
        False,
        False,
        True,
    ],
    [
        "Bir zamanlar, altı yaşındayken, muhteşem bir resim gördüm",
        "tr",
        "examples/female.wav",
        None,
        False,
        False,
        False,
        True,
    ],
    [
        "Когда мне было шесть лет, я увидел однажды удивительную картинку",
        "ru",
        "examples/female.wav",
        None,
       False,
        False,
        False,
        True,
    ],
    [
        "Toen ik een jaar of zes was, zag ik op een keer een prachtige plaat",
        "nl",
        "examples/male.wav",
        None,
       False,
        False,
        False,
        True,
    ],
    [
        "Když mi bylo šest let, viděl jsem jednou nádherný obrázek",
        "cs",
        "examples/female.wav",
        None,
       False,
        False,
        False,
        True,
    ],
    [
        "当我还只有六岁的时候, 看到了一副精彩的插画",
        "zh-cn",
        "examples/female.wav",
        None,
       False,
        False,
        False,
        True,
    ],
]



gr.Interface(
    fn=predict,
    inputs=[
        gr.Textbox(
            label="Text Prompt",
            info="One or two sentences at a time is better. Up to 200 text characters.",
            value="Hi there, I'm your new voice clone. Try your best to upload quality audio",
        ),
        gr.Dropdown(
            label="Language",
            info="Select an output language for the synthesised speech",
            choices=[
                "en",
                "es",
                "fr",
                "de",
                "it",
                "pt",
                "pl",
                "tr",
                "ru",
                "nl",
                "cs",
                "ar",
                "zh-cn",
            ],
            max_choices=1,
            value="en",
        ),
        gr.Audio(
            label="Reference Audio",
            info="Click on the ✎ button to upload your own target speaker audio",
            type="filepath",
            value="examples/female.wav",
        ),
        gr.Audio(source="microphone",
                 type="filepath",
                 info="Use your microphone to record audio",
                 label="Use Microphone for Reference"),
        gr.Checkbox(label="Use Microphone",
                    value=False,
                    info="Notice: Microphone input may not work properly under traffic",),
        gr.Checkbox(label="Cleanup Reference Voice",
                    value=False,
                    info="This check can improve output if your microphone or reference voice is noisy",
                    ),
        gr.Checkbox(label="Do not use language auto-detect",
                    value=False,
                    info="Check to disable language auto-detection",),
        gr.Checkbox(
            label="Agree",
            value=False,
            info="I agree to the terms of the Coqui Public Model License at https://coqui.ai/cpml",
        ),

        
    ],
    outputs=[
        gr.Video(label="Waveform Visual"),
        gr.Audio(label="Synthesised Audio",autoplay=True),
        gr.Text(label="Metrics"),
        gr.Audio(label="Reference Audio Used"),
    ],
    title=title,
    description=description,
    article=article,
    examples=examples,
).queue().launch(debug=True,show_api=False)