BilalSardar commited on
Commit
490705b
1 Parent(s): 5613cf0

Upload 10 files

Browse files
.gitattributes CHANGED
@@ -31,3 +31,5 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
31
  *.zip filter=lfs diff=lfs merge=lfs -text
32
  *.zst filter=lfs diff=lfs merge=lfs -text
33
  *tfevents* filter=lfs diff=lfs merge=lfs -text
 
 
 
31
  *.zip filter=lfs diff=lfs merge=lfs -text
32
  *.zst filter=lfs diff=lfs merge=lfs -text
33
  *tfevents* filter=lfs diff=lfs merge=lfs -text
34
+ SE_checkpoint.pth.tar filter=lfs diff=lfs merge=lfs -text
35
+ best_model_latest.pth.tar filter=lfs diff=lfs merge=lfs -text
SE_checkpoint.pth.tar ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8f96efb20cbeeefd81fd8336d7f0155bf8902f82f9474e58ccb19d9e12345172
3
+ size 44610930
app.py ADDED
@@ -0,0 +1,158 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ from turtle import title
2
+ import gradio as gr
3
+
4
+ import git
5
+ import os
6
+ os.system('git clone https://github.com/Edresson/Coqui-TTS -b multilingual-torchaudio-SE TTS')
7
+ os.system('pip install -q -e TTS/')
8
+ os.system('pip install -q torchaudio==0.9.0')
9
+
10
+ import sys
11
+ TTS_PATH = "TTS/"
12
+
13
+ # add libraries into environment
14
+ sys.path.append(TTS_PATH) # set this if TTS is not installed globally
15
+
16
+ import os
17
+ import string
18
+ import time
19
+ import argparse
20
+ import json
21
+
22
+ import numpy as np
23
+ import IPython
24
+ from IPython.display import Audio
25
+
26
+
27
+ import torch
28
+
29
+ from TTS.tts.utils.synthesis import synthesis
30
+ from TTS.tts.utils.text.symbols import make_symbols, phonemes, symbols
31
+ try:
32
+ from TTS.utils.audio import AudioProcessor
33
+ except:
34
+ from TTS.utils.audio import AudioProcessor
35
+
36
+
37
+ from TTS.tts.models import setup_model
38
+ from TTS.config import load_config
39
+ from TTS.tts.models.vits import *
40
+
41
+ OUT_PATH = 'out/'
42
+
43
+ # create output path
44
+ os.makedirs(OUT_PATH, exist_ok=True)
45
+
46
+ # model vars
47
+ MODEL_PATH = '/home/user/app/best_model_latest.pth.tar'
48
+ CONFIG_PATH = '/home/user/app/config.json'
49
+ TTS_LANGUAGES = "/home/user/app/language_ids.json"
50
+ TTS_SPEAKERS = "/home/user/app/speakers.json"
51
+ USE_CUDA = torch.cuda.is_available()
52
+
53
+ # load the config
54
+ C = load_config(CONFIG_PATH)
55
+
56
+
57
+ # load the audio processor
58
+ ap = AudioProcessor(**C.audio)
59
+
60
+ speaker_embedding = None
61
+
62
+ C.model_args['d_vector_file'] = TTS_SPEAKERS
63
+ C.model_args['use_speaker_encoder_as_loss'] = False
64
+
65
+ model = setup_model(C)
66
+ model.language_manager.set_language_ids_from_file(TTS_LANGUAGES)
67
+ # print(model.language_manager.num_languages, model.embedded_language_dim)
68
+ # print(model.emb_l)
69
+ cp = torch.load(MODEL_PATH, map_location=torch.device('cpu'))
70
+ # remove speaker encoder
71
+ model_weights = cp['model'].copy()
72
+ for key in list(model_weights.keys()):
73
+ if "speaker_encoder" in key:
74
+ del model_weights[key]
75
+
76
+ model.load_state_dict(model_weights)
77
+
78
+
79
+ model.eval()
80
+
81
+ if USE_CUDA:
82
+ model = model.cuda()
83
+
84
+ # synthesize voice
85
+ use_griffin_lim = False
86
+
87
+ os.system('pip install -q pydub ffmpeg-normalize')
88
+
89
+ CONFIG_SE_PATH = "config_se.json"
90
+ CHECKPOINT_SE_PATH = "SE_checkpoint.pth.tar"
91
+
92
+ from TTS.tts.utils.speakers import SpeakerManager
93
+ from pydub import AudioSegment
94
+ import librosa
95
+
96
+ SE_speaker_manager = SpeakerManager(encoder_model_path=CHECKPOINT_SE_PATH, encoder_config_path=CONFIG_SE_PATH, use_cuda=USE_CUDA)
97
+
98
+ def compute_spec(ref_file):
99
+ y, sr = librosa.load(ref_file, sr=ap.sample_rate)
100
+ spec = ap.spectrogram(y)
101
+ spec = torch.FloatTensor(spec).unsqueeze(0)
102
+ return spec
103
+
104
+
105
+
106
+ def greet(Text,Voicetoclone):
107
+ text= "%s" % (Text)
108
+ reference_files= "%s" % (Voicetoclone)
109
+ print("path url")
110
+ print(Voicetoclone)
111
+ sample= str(Voicetoclone)
112
+ size= len(reference_files)*sys.getsizeof(reference_files)
113
+ size2= size / 1000000
114
+ if (size2 > 0.012) or len(text)>2000:
115
+ message="File is greater than 30mb or Text inserted is longer than 2000 characters. Please re-try with smaller sizes."
116
+ print(message)
117
+ raise SystemExit("File is greater than 30mb. Please re-try or Text inserted is longer than 2000 characters. Please re-try with smaller sizes.")
118
+ else:
119
+ os.system('ffmpeg-normalize $sample -nt rms -t=-27 -o $sample -ar 16000 -f')
120
+ reference_emb = SE_speaker_manager.compute_d_vector_from_clip(reference_files)
121
+ model.length_scale = 1 # scaler for the duration predictor. The larger it is, the slower the speech.
122
+ model.inference_noise_scale = 0.3 # defines the noise variance applied to the random z vector at inference.
123
+ model.inference_noise_scale_dp = 0.3 # defines the noise variance applied to the duration predictor z vector at inference.
124
+ text = text
125
+ model.language_manager.language_id_mapping
126
+ language_id = 0
127
+
128
+ print(" > text: {}".format(text))
129
+ wav, alignment, _, _ = synthesis(
130
+ model,
131
+ text,
132
+ C,
133
+ "cuda" in str(next(model.parameters()).device),
134
+ ap,
135
+ speaker_id=None,
136
+ d_vector=reference_emb,
137
+ style_wav=None,
138
+ language_id=language_id,
139
+ enable_eos_bos_chars=C.enable_eos_bos_chars,
140
+ use_griffin_lim=True,
141
+ do_trim_silence=False,
142
+ ).values()
143
+ print("Generated Audio")
144
+ IPython.display.display(Audio(wav, rate=ap.sample_rate))
145
+ file_name = text.replace(" ", "_")
146
+ file_name = file_name.translate(str.maketrans('', '', string.punctuation.replace('_', ''))) + '.wav'
147
+ out_path = os.path.join(OUT_PATH, file_name)
148
+ print(" > Saving output to {}".format(out_path))
149
+ ap.save_wav(wav, out_path)
150
+ return out_path
151
+
152
+ demo = gr.Interface(
153
+ fn=greet,
154
+ inputs=[gr.inputs.Textbox(label='What would you like the voice to say? (max. 2000 characters per request)'),gr.Audio(type="filepath", source="upload",label='Please upload a voice to clone (max. 30mb)')],
155
+ outputs="audio",
156
+ title="Bilal's Voice Cloning Tool"
157
+ )
158
+ demo.launch()
best_model_latest.pth.tar ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:017bfd8907c80bb5857d65d0223f0e4e4b9d699ef52e2a853d9cc7eb7e308cf0
3
+ size 379957289
config.json ADDED
@@ -0,0 +1,373 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "model": "vits",
3
+ "run_name": "vits_tts-portuguese",
4
+ "run_description": "",
5
+ "epochs": 1000,
6
+ "batch_size": 52,
7
+ "eval_batch_size": 52,
8
+ "mixed_precision": false,
9
+ "scheduler_after_epoch": true,
10
+ "run_eval": true,
11
+ "test_delay_epochs": -1,
12
+ "print_eval": true,
13
+ "dashboard_logger": "tensorboard",
14
+ "print_step": 25,
15
+ "plot_step": 100,
16
+ "model_param_stats": false,
17
+ "project_name": null,
18
+ "log_model_step": 10000,
19
+ "wandb_entity": null,
20
+ "save_step": 10000,
21
+ "checkpoint": true,
22
+ "keep_all_best": false,
23
+ "keep_after": 10000,
24
+ "num_loader_workers": 4,
25
+ "num_eval_loader_workers": 4,
26
+ "use_noise_augment": false,
27
+ "use_language_weighted_sampler": true,
28
+ "output_path": "../checkpoints/VITS-multilingual/VITS_fixes/new/new-SE/use_noise_aument_false/xlarge-ZS-PT-VCTK/pt-en+LibriTTS-fr/speaker_encoder_as_loss_9_alpha/mixed-p-false-bug-SDP-fixed/",
29
+ "distributed_backend": "nccl",
30
+ "distributed_url": "tcp://localhost:54321",
31
+ "audio": {
32
+ "fft_size": 1024,
33
+ "win_length": 1024,
34
+ "hop_length": 256,
35
+ "frame_shift_ms": null,
36
+ "frame_length_ms": null,
37
+ "stft_pad_mode": "reflect",
38
+ "sample_rate": 16000,
39
+ "resample": false,
40
+ "preemphasis": 0.0,
41
+ "ref_level_db": 20,
42
+ "do_sound_norm": false,
43
+ "log_func": "np.log",
44
+ "do_trim_silence": true,
45
+ "trim_db": 45,
46
+ "power": 1.5,
47
+ "griffin_lim_iters": 60,
48
+ "num_mels": 80,
49
+ "mel_fmin": 0.0,
50
+ "mel_fmax": null,
51
+ "spec_gain": 1,
52
+ "do_amp_to_db_linear": false,
53
+ "do_amp_to_db_mel": true,
54
+ "signal_norm": false,
55
+ "min_level_db": -100,
56
+ "symmetric_norm": true,
57
+ "max_norm": 4.0,
58
+ "clip_norm": true,
59
+ "stats_path": null
60
+ },
61
+ "use_phonemes": false,
62
+ "use_espeak_phonemes": false,
63
+ "phoneme_language": "pt-br",
64
+ "compute_input_seq_cache": false,
65
+ "text_cleaner": "multilingual_cleaners",
66
+ "enable_eos_bos_chars": false,
67
+ "test_sentences_file": "",
68
+ "phoneme_cache_path": null,
69
+ "characters": {
70
+ "pad": "_",
71
+ "eos": "&",
72
+ "bos": "*",
73
+ "characters": "ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz\u00af\u00b7\u00df\u00e0\u00e1\u00e2\u00e3\u00e4\u00e6\u00e7\u00e8\u00e9\u00ea\u00eb\u00ec\u00ed\u00ee\u00ef\u00f1\u00f2\u00f3\u00f4\u00f5\u00f6\u00f9\u00fa\u00fb\u00fc\u00ff\u0101\u0105\u0107\u0113\u0119\u011b\u012b\u0131\u0142\u0144\u014d\u0151\u0153\u015b\u016b\u0171\u017a\u017c\u01ce\u01d0\u01d2\u01d4\u0430\u0431\u0432\u0433\u0434\u0435\u0436\u0437\u0438\u0439\u043a\u043b\u043c\u043d\u043e\u043f\u0440\u0441\u0442\u0443\u0444\u0445\u0446\u0447\u0448\u0449\u044a\u044b\u044c\u044d\u044e\u044f\u0451\u0454\u0456\u0457\u0491\u2013!'(),-.:;? ",
74
+ "punctuations": "!'(),-.:;? ",
75
+ "phonemes": "iy\u0268\u0289\u026fu\u026a\u028f\u028ae\u00f8\u0258\u0259\u0275\u0264o\u025b\u0153\u025c\u025e\u028c\u0254\u00e6\u0250a\u0276\u0251\u0252\u1d7b\u0298\u0253\u01c0\u0257\u01c3\u0284\u01c2\u0260\u01c1\u029bpbtd\u0288\u0256c\u025fk\u0261q\u0262\u0294\u0274\u014b\u0272\u0273n\u0271m\u0299r\u0280\u2c71\u027e\u027d\u0278\u03b2fv\u03b8\u00f0sz\u0283\u0292\u0282\u0290\u00e7\u029dx\u0263\u03c7\u0281\u0127\u0295h\u0266\u026c\u026e\u028b\u0279\u027bj\u0270l\u026d\u028e\u029f\u02c8\u02cc\u02d0\u02d1\u028dw\u0265\u029c\u02a2\u02a1\u0255\u0291\u027a\u0267\u025a\u02de\u026b'\u0303' ",
76
+ "unique": true
77
+ },
78
+ "batch_group_size": 0,
79
+ "loss_masking": null,
80
+ "min_seq_len": 90,
81
+ "max_seq_len": 270,
82
+ "compute_f0": false,
83
+ "compute_linear_spec": true,
84
+ "add_blank": true,
85
+ "datasets": [
86
+ {
87
+ "name": "vctk",
88
+ "path": "../../datasets/VCTK-Corpus-removed-silence_16Khz/",
89
+ "meta_file_train": null,
90
+ "ununsed_speakers": [
91
+ "p225",
92
+ "p234",
93
+ "p238",
94
+ "p245",
95
+ "p248",
96
+ "p261",
97
+ "p294",
98
+ "p302",
99
+ "p326",
100
+ "p335",
101
+ "p347"
102
+ ],
103
+ "language": "en",
104
+ "meta_file_val": null,
105
+ "meta_file_attn_mask": ""
106
+ },
107
+ {
108
+ "name": "libri_tts",
109
+ "path": "../../datasets/LibriTTS/LibriTTS/dataset-preprocessed-clean-100-and-360/dataset-22k/",
110
+ "meta_file_train": "metadata_all.csv",
111
+ "ununsed_speakers": null,
112
+ "language": "en",
113
+ "meta_file_val": "dev-clean_500.csv",
114
+ "meta_file_attn_mask": ""
115
+ },
116
+ {
117
+ "name": "brspeech",
118
+ "path": "../../datasets/TTS-Portuguese-Corpus_16khz/",
119
+ "meta_file_train": "train_TTS-Portuguese_Corpus_metadata.csv",
120
+ "ununsed_speakers": null,
121
+ "language": "pt-br",
122
+ "meta_file_val": "eval_TTS-Portuguese_Corpus_metadata.csv",
123
+ "meta_file_attn_mask": ""
124
+ },
125
+ {
126
+ "name": "mailabs",
127
+ "path": "../../datasets/M-AILABS/fr_FR",
128
+ "meta_file_train": "",
129
+ "ununsed_speakers": null,
130
+ "language": "fr-fr",
131
+ "meta_file_val": null,
132
+ "meta_file_attn_mask": null
133
+ }
134
+ ],
135
+ "optimizer": "AdamW",
136
+ "optimizer_params": {
137
+ "betas": [
138
+ 0.8,
139
+ 0.99
140
+ ],
141
+ "eps": 1e-09,
142
+ "weight_decay": 0.01
143
+ },
144
+ "lr_scheduler": "",
145
+ "lr_scheduler_params": null,
146
+ "test_sentences": [
147
+ [
148
+ "It took me quite a long time to develop a voice, and now that I have it I'm not going to be silent.",
149
+ "VCTK_p225",
150
+ null,
151
+ "en"
152
+ ],
153
+ [
154
+ "It took me quite a long time to develop a voice, and now that I have it I'm not going to be silent.",
155
+ "ED",
156
+ null,
157
+ "en"
158
+ ],
159
+ [
160
+ "It took me quite a long time to develop a voice, and now that I have it I'm not going to be silent.",
161
+ "bernard",
162
+ null,
163
+ "en"
164
+ ],
165
+ [
166
+ "This cake is great. It's so delicious and moist.",
167
+ "VCTK_p234",
168
+ null,
169
+ "en"
170
+ ],
171
+ [
172
+ "This cake is great. It's so delicious and moist.",
173
+ "ED",
174
+ null,
175
+ "en"
176
+ ],
177
+ [
178
+ "This cake is great. It's so delicious and moist.",
179
+ "ezwa",
180
+ null,
181
+ "en"
182
+ ],
183
+ [
184
+ "Hoje \u00e9 fundamental encontrar a raz\u00e3o da exist\u00eancia humana.",
185
+ "ED",
186
+ null,
187
+ "pt-br"
188
+ ],
189
+ [
190
+ "Hoje \u00e9 fundamental encontrar a raz\u00e3o da exist\u00eancia humana.",
191
+ "VCTK_p238",
192
+ null,
193
+ "pt-br"
194
+ ],
195
+ [
196
+ "Hoje \u00e9 fundamental encontrar a raz\u00e3o da exist\u00eancia humana.",
197
+ "gilles_g_le_blanc",
198
+ null,
199
+ "pt-br"
200
+ ],
201
+ [
202
+ "Em muitas cidades a popula\u00e7\u00e3o est\u00e1 diminuindo.",
203
+ "ED",
204
+ null,
205
+ "pt-br"
206
+ ],
207
+ [
208
+ "Em muitas cidades a popula\u00e7\u00e3o est\u00e1 diminuindo.",
209
+ "VCTK_p245",
210
+ null,
211
+ "pt-br"
212
+ ],
213
+ [
214
+ "Em muitas cidades a popula\u00e7\u00e3o est\u00e1 diminuindo.",
215
+ "nadine_eckert_boulet",
216
+ null,
217
+ "pt-br"
218
+ ],
219
+ [
220
+ "Il m'a fallu beaucoup de temps pour d\u00e9velopper une voix, et maintenant que je l'ai, je ne vais pas me taire.",
221
+ "VCTK_p245",
222
+ null,
223
+ "fr-fr"
224
+ ],
225
+ [
226
+ "Il m'a fallu beaucoup de temps pour d\u00e9velopper une voix, et maintenant que je l'ai, je ne vais pas me taire.",
227
+ "ED",
228
+ null,
229
+ "fr-fr"
230
+ ],
231
+ [
232
+ "Il m'a fallu beaucoup de temps pour d\u00e9velopper une voix, et maintenant que je l'ai, je ne vais pas me taire.",
233
+ "ezwa",
234
+ null,
235
+ "fr-fr"
236
+ ],
237
+ [
238
+ "Il m'a fallu beaucoup de temps pour d\u00e9velopper une voix, et maintenant que je l'ai, je ne vais pas me taire.",
239
+ "bernard",
240
+ null,
241
+ "fr-fr"
242
+ ],
243
+ [
244
+ "Il m'a fallu beaucoup de temps pour d\u00e9velopper une voix, et maintenant que je l'ai, je ne vais pas me taire.",
245
+ "gilles_g_le_blanc",
246
+ null,
247
+ "fr-fr"
248
+ ],
249
+ [
250
+ "Il m'a fallu beaucoup de temps pour d\u00e9velopper une voix, et maintenant que je l'ai, je ne vais pas me taire.",
251
+ "nadine_eckert_boulet",
252
+ null,
253
+ "fr-fr"
254
+ ],
255
+ [
256
+ "Il m'a fallu beaucoup de temps pour d\u00e9velopper une voix, et maintenant que je l'ai, je ne vais pas me taire.",
257
+ "zeckou",
258
+ null,
259
+ "fr-fr"
260
+ ]
261
+ ],
262
+ "use_speaker_embedding": true,
263
+ "use_d_vector_file": true,
264
+ "d_vector_dim": 512,
265
+ "model_args": {
266
+ "num_chars": 165,
267
+ "out_channels": 513,
268
+ "spec_segment_size": 62,
269
+ "hidden_channels": 192,
270
+ "hidden_channels_ffn_text_encoder": 768,
271
+ "num_heads_text_encoder": 2,
272
+ "num_layers_text_encoder": 10,
273
+ "kernel_size_text_encoder": 3,
274
+ "dropout_p_text_encoder": 0.1,
275
+ "dropout_p_duration_predictor": 0.5,
276
+ "kernel_size_posterior_encoder": 5,
277
+ "dilation_rate_posterior_encoder": 1,
278
+ "num_layers_posterior_encoder": 16,
279
+ "kernel_size_flow": 5,
280
+ "dilation_rate_flow": 1,
281
+ "num_layers_flow": 4,
282
+ "resblock_type_decoder": 1,
283
+ "resblock_kernel_sizes_decoder": [
284
+ 3,
285
+ 7,
286
+ 11
287
+ ],
288
+ "resblock_dilation_sizes_decoder": [
289
+ [
290
+ 1,
291
+ 3,
292
+ 5
293
+ ],
294
+ [
295
+ 1,
296
+ 3,
297
+ 5
298
+ ],
299
+ [
300
+ 1,
301
+ 3,
302
+ 5
303
+ ]
304
+ ],
305
+ "upsample_rates_decoder": [
306
+ 8,
307
+ 8,
308
+ 2,
309
+ 2
310
+ ],
311
+ "upsample_initial_channel_decoder": 512,
312
+ "upsample_kernel_sizes_decoder": [
313
+ 16,
314
+ 16,
315
+ 4,
316
+ 4
317
+ ],
318
+ "use_sdp": true,
319
+ "noise_scale": 1.0,
320
+ "inference_noise_scale": 0.667,
321
+ "length_scale": 1,
322
+ "noise_scale_dp": 1.0,
323
+ "inference_noise_scale_dp": 0.8,
324
+ "max_inference_len": null,
325
+ "init_discriminator": true,
326
+ "use_spectral_norm_disriminator": false,
327
+ "use_speaker_embedding": true,
328
+ "num_speakers": 1244,
329
+ "speakers_file": null,
330
+ "d_vector_file": "../speaker_embeddings/new-SE/VCTK-LibriTTS+TTS-PT+MAILABS-FR/speakers.json",
331
+ "speaker_embedding_channels": 512,
332
+ "use_d_vector_file": true,
333
+ "d_vector_dim": 512,
334
+ "detach_dp_input": true,
335
+ "use_language_embedding": true,
336
+ "embedded_language_dim": 4,
337
+ "num_languages": 3,
338
+ "use_speaker_encoder_as_loss": true,
339
+ "speaker_encoder_config_path": "../checkpoints/Speaker_Encoder/Resnet-original-paper/config.json",
340
+ "speaker_encoder_model_path": "../checkpoints/Speaker_Encoder/Resnet-original-paper/converted_checkpoint.pth.tar",
341
+ "fine_tuning_mode": 0,
342
+ "freeze_encoder": false,
343
+ "freeze_DP": false,
344
+ "freeze_PE": false,
345
+ "freeze_flow_decoder": false,
346
+ "freeze_waveform_decoder": false
347
+ },
348
+ "grad_clip": [
349
+ 5.0,
350
+ 5.0
351
+ ],
352
+ "lr_gen": 0.0002,
353
+ "lr_disc": 0.0002,
354
+ "lr_scheduler_gen": "ExponentialLR",
355
+ "lr_scheduler_gen_params": {
356
+ "gamma": 0.999875,
357
+ "last_epoch": -1
358
+ },
359
+ "lr_scheduler_disc": "ExponentialLR",
360
+ "lr_scheduler_disc_params": {
361
+ "gamma": 0.999875,
362
+ "last_epoch": -1
363
+ },
364
+ "kl_loss_alpha": 1.0,
365
+ "disc_loss_alpha": 1.0,
366
+ "gen_loss_alpha": 1.0,
367
+ "feat_loss_alpha": 1.0,
368
+ "mel_loss_alpha": 45.0,
369
+ "dur_loss_alpha": 1.0,
370
+ "speaker_encoder_loss_alpha": 9.0,
371
+ "return_wav": true,
372
+ "r": 1
373
+ }
config_se.json ADDED
@@ -0,0 +1,119 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "model": "speaker_encoder",
3
+ "run_name": "speaker_encoder",
4
+ "run_description": "resnet speaker encoder trained with commonvoice all languages dev and train, Voxceleb 1 dev and Voxceleb 2 dev",
5
+ "epochs": 100000,
6
+ "batch_size": null,
7
+ "eval_batch_size": null,
8
+ "mixed_precision": false,
9
+ "run_eval": true,
10
+ "test_delay_epochs": 0,
11
+ "print_eval": false,
12
+ "print_step": 50,
13
+ "tb_plot_step": 100,
14
+ "tb_model_param_stats": false,
15
+ "save_step": 1000,
16
+ "checkpoint": true,
17
+ "keep_all_best": false,
18
+ "keep_after": 10000,
19
+ "num_loader_workers": 8,
20
+ "num_val_loader_workers": 0,
21
+ "use_noise_augment": false,
22
+ "output_path": "../checkpoints/speaker_encoder/language_balanced/normalized/angleproto-4-samples-by-speakers/",
23
+ "distributed_backend": "nccl",
24
+ "distributed_url": "tcp://localhost:54321",
25
+ "audio": {
26
+ "fft_size": 512,
27
+ "win_length": 400,
28
+ "hop_length": 160,
29
+ "frame_shift_ms": null,
30
+ "frame_length_ms": null,
31
+ "stft_pad_mode": "reflect",
32
+ "sample_rate": 16000,
33
+ "resample": false,
34
+ "preemphasis": 0.97,
35
+ "ref_level_db": 20,
36
+ "do_sound_norm": false,
37
+ "do_trim_silence": false,
38
+ "trim_db": 60,
39
+ "power": 1.5,
40
+ "griffin_lim_iters": 60,
41
+ "num_mels": 64,
42
+ "mel_fmin": 0.0,
43
+ "mel_fmax": 8000.0,
44
+ "spec_gain": 20,
45
+ "signal_norm": false,
46
+ "min_level_db": -100,
47
+ "symmetric_norm": false,
48
+ "max_norm": 4.0,
49
+ "clip_norm": false,
50
+ "stats_path": null
51
+ },
52
+ "datasets": [
53
+ {
54
+ "name": "voxceleb2",
55
+ "path": "/workspace/scratch/ecasanova/datasets/VoxCeleb/vox2_dev_aac/",
56
+ "meta_file_train": null,
57
+ "ununsed_speakers": null,
58
+ "meta_file_val": null,
59
+ "meta_file_attn_mask": "",
60
+ "language": "voxceleb"
61
+ }
62
+ ],
63
+ "model_params": {
64
+ "model_name": "resnet",
65
+ "input_dim": 64,
66
+ "use_torch_spec": true,
67
+ "log_input": true,
68
+ "proj_dim": 512
69
+ },
70
+ "audio_augmentation": {
71
+ "p": 0.5,
72
+ "rir": {
73
+ "rir_path": "/workspace/store/ecasanova/ComParE/RIRS_NOISES/simulated_rirs/",
74
+ "conv_mode": "full"
75
+ },
76
+ "additive": {
77
+ "sounds_path": "/workspace/store/ecasanova/ComParE/musan/",
78
+ "speech": {
79
+ "min_snr_in_db": 13,
80
+ "max_snr_in_db": 20,
81
+ "min_num_noises": 1,
82
+ "max_num_noises": 1
83
+ },
84
+ "noise": {
85
+ "min_snr_in_db": 0,
86
+ "max_snr_in_db": 15,
87
+ "min_num_noises": 1,
88
+ "max_num_noises": 1
89
+ },
90
+ "music": {
91
+ "min_snr_in_db": 5,
92
+ "max_snr_in_db": 15,
93
+ "min_num_noises": 1,
94
+ "max_num_noises": 1
95
+ }
96
+ },
97
+ "gaussian": {
98
+ "p": 0.0,
99
+ "min_amplitude": 0.0,
100
+ "max_amplitude": 1e-05
101
+ }
102
+ },
103
+ "storage": {
104
+ "sample_from_storage_p": 0.5,
105
+ "storage_size": 40
106
+ },
107
+ "max_train_step": 1000000,
108
+ "loss": "angleproto",
109
+ "grad_clip": 3.0,
110
+ "lr": 0.0001,
111
+ "lr_decay": false,
112
+ "warmup_steps": 4000,
113
+ "wd": 1e-06,
114
+ "steps_plot_stats": 100,
115
+ "num_speakers_in_batch": 100,
116
+ "num_utters_per_speaker": 4,
117
+ "skip_speakers": true,
118
+ "voice_len": 2.0
119
+ }
cv-speakers-pt+en-m-f.json ADDED
The diff for this file is too large to render. See raw diff
 
errormessage.wav ADDED
Binary file (889 kB). View file
 
language_ids.json ADDED
@@ -0,0 +1,5 @@
 
 
 
 
 
 
1
+ {
2
+ "en": 0,
3
+ "fr-fr": 1,
4
+ "pt-br": 2
5
+ }
requirements.txt ADDED
@@ -0,0 +1,2 @@
 
 
 
1
+ TTS
2
+ torchaudio==0.9.0
speakers.json ADDED
The diff for this file is too large to render. See raw diff