Hugging Face's logo

verstar
/
TechSinger

Text-to-Speech
Model card Files
xet
Community
1
3v324v23 commited on
Commit
e23c915
·
1 Parent(s): 43dc628

Upload bilingual

Browse files
Files changed (3) hide show
  1. bilingual/stage1/model_ckpt_steps_200000.ckpt +3 -0
  2. bilingual/stage2/config.yaml +398 -0
  3. bilingual/stage2/model_ckpt_steps_160000.ckpt +3 -0
bilingual/stage1/model_ckpt_steps_200000.ckpt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0cd5e5f2410c4b0bfab2ac8593faeb9e28337d318ba6c75e33fc4c8d3f0f0773
3
+ size 304957924
bilingual/stage2/config.yaml ADDED
@@ -0,0 +1,398 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ K_step: 100
2
+ accumulate_grad_batches: 1
3
+ amp: false
4
+ audio_num_mel_bins: 80
5
+ audio_sample_rate: 48000
6
+ base_config:
7
+ - egs/stage1.yaml
8
+ binarization_args:
9
+ min_sil_duration: 0.1
10
+ reset_phone_dict: true
11
+ reset_word_dict: true
12
+ shuffle: false
13
+ test_range:
14
+ - 0
15
+ - 100
16
+ train_range:
17
+ - 200
18
+ - -1
19
+ trim_eos_bos: false
20
+ trim_sil: false
21
+ valid_range:
22
+ - 100
23
+ - 200
24
+ with_align: true
25
+ with_f0: true
26
+ with_f0cwt: false
27
+ with_linear: false
28
+ with_spk_embed: false
29
+ with_spk_id: true
30
+ with_txt: true
31
+ with_wav: true
32
+ with_word: true
33
+ binarizer_cls: data_gen.tech_binarizer.TechBinarizer
34
+ binary_data_dir: data/binary/bilingual
35
+ cfg_scale: 1.0
36
+ check_val_every_n_epoch: 10
37
+ clip_grad_norm: 1
38
+ clip_grad_value: 0
39
+ conv_use_pos: false
40
+ debug: false
41
+ dec_dilations:
42
+ - 1
43
+ - 1
44
+ - 1
45
+ - 1
46
+ dec_ffn_kernel_size: 9
47
+ dec_inp_add_noise: false
48
+ dec_kernel_size: 5
49
+ dec_layers: 4
50
+ dec_post_net_kernel: 3
51
+ decay_steps: 50000
52
+ decoder_rnn_dim: 0
53
+ decoder_type: conv
54
+ diff_decoder_type: wavenet
55
+ diff_loss_type: l1
56
+ dilation_cycle_length: 4
57
+ drop_tech_prob: 0.2
58
+ dropout: 0.0
59
+ ds_workers: 4
60
+ dur_level: ph
61
+ dur_loss: mse
62
+ dur_predictor_kernel: 3
63
+ dur_predictor_layers: 2
64
+ enc_dec_norm: ln
65
+ enc_dilations:
66
+ - 1
67
+ - 1
68
+ - 1
69
+ - 1
70
+ enc_ffn_kernel_size: 9
71
+ enc_kernel_size: 5
72
+ enc_layers: 4
73
+ enc_post_net_kernel: 3
74
+ enc_pre_ln: true
75
+ enc_prenet: true
76
+ encoder_K: 8
77
+ encoder_type: rel_fft
78
+ endless_ds: true
79
+ eval_max_batches: -1
80
+ f0_K_step: 100
81
+ f0_dilation_cycle_length: 4
82
+ f0_gen: flow
83
+ f0_infer_with_ref: false
84
+ f0_max: 1000
85
+ f0_max_beta: 0.06
86
+ f0_min: 50
87
+ f0_residual_channels: 192
88
+ f0_residual_layers: 10
89
+ f0_sample_clip: true
90
+ f0_timesteps: 1000
91
+ ffn_act: gelu
92
+ ffn_hidden_size: 1024
93
+ fft_size: 1024
94
+ flow_decoder_type: wavenet
95
+ flow_loss_type: l1
96
+ flow_qsample: direct
97
+ fmax: 24000
98
+ fmin: 20
99
+ frames_multiple: 1
100
+ fs2_ckpt_dir: checkpoints/stage1
101
+ gaussian_start: false
102
+ gen_dir_name: ''
103
+ griffin_lim_iters: 30
104
+ hidden_size: 256
105
+ hop_size: 256
106
+ infer: false
107
+ keep_bins: 80
108
+ lambda_commit: 0.25
109
+ lambda_energy: 0.1
110
+ lambda_f0: 1.0
111
+ lambda_ph_dur: 1.0
112
+ lambda_sent_dur: 0.0
113
+ lambda_uv: 1.0
114
+ lambda_word_dur: 0.0
115
+ layers_in_block: 2
116
+ load_ckpt: ''
117
+ loud_norm: false
118
+ lr: 0.001
119
+ max_beta: 0.06
120
+ max_epochs: 1000
121
+ max_frames: 3000
122
+ max_input_tokens: 1550
123
+ max_sentences: 16
124
+ max_tokens: 24000
125
+ max_updates: 160000
126
+ max_valid_sentences: 1
127
+ max_valid_tokens: 60000
128
+ mel_loss_scale: direct
129
+ mel_losses: l1:0.5|ssim:0.5
130
+ mel_vmax: 1.5
131
+ mel_vmin: -6
132
+ min_frames: 0
133
+ num_ckpt_keep: 3
134
+ num_heads: 2
135
+ num_sanity_val_steps: 0
136
+ num_spk: 160
137
+ num_valid_plots: 10
138
+ optimizer_adam_beta1: 0.9
139
+ optimizer_adam_beta2: 0.98
140
+ out_wav_norm: false
141
+ pitch_extractor: parselmouth
142
+ pitch_key: pitch
143
+ pitch_type: frame
144
+ predictor_dropout: 0.0
145
+ predictor_grad: 1.0
146
+ predictor_hidden: -1
147
+ predictor_kernel: 5
148
+ predictor_layers: 5
149
+ preprocess_args:
150
+ add_eos_bos: true
151
+ mfa_group_shuffle: false
152
+ mfa_offset: 0.02
153
+ nsample_per_mfa_group: 1000
154
+ reset_phone_dict: true
155
+ reset_word_dict: true
156
+ save_sil_mask: true
157
+ txt_processor: en
158
+ use_mfa: true
159
+ vad_max_silence_length: 12
160
+ wav_processors: []
161
+ with_phsep: true
162
+ preprocess_cls: ''
163
+ print_nan_grads: false
164
+ processed_data_dir: data/processed/bilingual
165
+ profile_infer: false
166
+ raw_data_dir: ''
167
+ ref_norm_layer: bn
168
+ rename_tmux: true
169
+ residual_channels: 256
170
+ residual_layers: 20
171
+ resume_from_checkpoint: 0
172
+ save_best: false
173
+ save_codes:
174
+ - tasks
175
+ - modules
176
+ - egs
177
+ save_f0: false
178
+ save_gt: true
179
+ schedule_type: linear
180
+ scheduler: warmup
181
+ seed: 1234
182
+ sort_by_len: true
183
+ spec_max:
184
+ - 0.03640973940491676
185
+ - 0.039425432682037354
186
+ - 0.29524752497673035
187
+ - 0.45784831047058105
188
+ - 0.48333120346069336
189
+ - 0.5335848927497864
190
+ - 0.6071611046791077
191
+ - 0.5474293828010559
192
+ - 0.6076506972312927
193
+ - 0.5390501022338867
194
+ - 0.5743886232376099
195
+ - 0.485751211643219
196
+ - 0.4248744249343872
197
+ - 0.4843744933605194
198
+ - 0.43331536650657654
199
+ - 0.5356124639511108
200
+ - 0.4875929355621338
201
+ - 0.48614853620529175
202
+ - 0.44228559732437134
203
+ - 0.5027499198913574
204
+ - 0.6554337739944458
205
+ - 0.3469322919845581
206
+ - 0.33981558680534363
207
+ - 0.37933868169784546
208
+ - 0.34751009941101074
209
+ - 0.22094282507896423
210
+ - 0.252963662147522
211
+ - 0.18274202942848206
212
+ - 0.1976650059223175
213
+ - 0.1770155429840088
214
+ - 0.18206502497196198
215
+ - 0.1002601608633995
216
+ - 0.18640224635601044
217
+ - 0.27240633964538574
218
+ - 0.04153885692358017
219
+ - -0.010289354249835014
220
+ - -0.012929759919643402
221
+ - 0.035185474902391434
222
+ - 0.18124309182167053
223
+ - -0.14512233436107635
224
+ - -0.1778590828180313
225
+ - -0.20491982996463776
226
+ - -0.30119436979293823
227
+ - -0.1735714226961136
228
+ - -0.1039585992693901
229
+ - -0.177497997879982
230
+ - -0.28803232312202454
231
+ - -0.24049188196659088
232
+ - -0.4682924747467041
233
+ - -0.5791841745376587
234
+ - -0.5170156955718994
235
+ - -0.6380605697631836
236
+ - -0.7147259712219238
237
+ - -0.6607836484909058
238
+ - -0.7288452982902527
239
+ - -0.6338580250740051
240
+ - -0.7092624306678772
241
+ - -0.8101216554641724
242
+ - -0.7633087038993835
243
+ - -0.8251329660415649
244
+ - -0.6936700940132141
245
+ - -0.5180960297584534
246
+ - -0.7972619533538818
247
+ - -0.807314932346344
248
+ - -0.7151175737380981
249
+ - -0.7785399556159973
250
+ - -0.8709449768066406
251
+ - -0.8360402584075928
252
+ - -0.8253681659698486
253
+ - -0.9778416156768799
254
+ - -1.12929368019104
255
+ - -1.3274869918823242
256
+ - -1.3071579933166504
257
+ - -1.5234452486038208
258
+ - -1.6191706657409668
259
+ - -1.708594799041748
260
+ - -1.8246771097183228
261
+ - -1.9193823337554932
262
+ - -2.1361801624298096
263
+ - -2.3829283714294434
264
+ spec_min:
265
+ - -6.0
266
+ - -6.0
267
+ - -6.0
268
+ - -6.0
269
+ - -6.0
270
+ - -6.0
271
+ - -6.0
272
+ - -6.0
273
+ - -6.0
274
+ - -6.0
275
+ - -6.0
276
+ - -6.0
277
+ - -6.0
278
+ - -6.0
279
+ - -6.0
280
+ - -6.0
281
+ - -6.0
282
+ - -6.0
283
+ - -6.0
284
+ - -6.0
285
+ - -6.0
286
+ - -6.0
287
+ - -6.0
288
+ - -6.0
289
+ - -6.0
290
+ - -6.0
291
+ - -6.0
292
+ - -6.0
293
+ - -6.0
294
+ - -6.0
295
+ - -6.0
296
+ - -6.0
297
+ - -6.0
298
+ - -6.0
299
+ - -6.0
300
+ - -6.0
301
+ - -6.0
302
+ - -6.0
303
+ - -6.0
304
+ - -6.0
305
+ - -6.0
306
+ - -6.0
307
+ - -6.0
308
+ - -6.0
309
+ - -6.0
310
+ - -6.0
311
+ - -6.0
312
+ - -6.0
313
+ - -6.0
314
+ - -6.0
315
+ - -6.0
316
+ - -6.0
317
+ - -6.0
318
+ - -6.0
319
+ - -6.0
320
+ - -6.0
321
+ - -6.0
322
+ - -6.0
323
+ - -6.0
324
+ - -6.0
325
+ - -6.0
326
+ - -6.0
327
+ - -6.0
328
+ - -6.0
329
+ - -6.0
330
+ - -6.0
331
+ - -6.0
332
+ - -6.0
333
+ - -6.0
334
+ - -6.0
335
+ - -6.0
336
+ - -6.0
337
+ - -6.0
338
+ - -6.0
339
+ - -6.0
340
+ - -6.0
341
+ - -6.0
342
+ - -6.0
343
+ - -6.0
344
+ - -6.0
345
+ task_cls: tasks.TechSinger.techsinger.RFPostnetTask
346
+ tb_log_interval: 100
347
+ test_ids: []
348
+ test_input_yaml: ''
349
+ test_num: 100
350
+ test_prefixes:
351
+ - "Chinese#ZH-Alto-1#Breathy#\u4E0D\u518D\u89C1"
352
+ - "Chinese#ZH-Tenor-1#Pharyngeal#\u4E0D\u4E3A\u8C01\u800C\u4F5C\u7684\u6B4C"
353
+ - "Chinese#ZH-Alto-1#Vibrato#\u4E0D\u518D\u89C1"
354
+ - "Chinese#ZH-Tenor-1#Glissando#\u4E0D\u67D3"
355
+ - "Chinese#ZH-Alto-1#Mixed_Voice_and_Falsetto#\u4E00\u6B21\u5C31\u597D"
356
+ - English#EN-Alto-1#Breathy#all is found
357
+ - English#EN-Alto-1#Pharyngeal#beauty and the beast
358
+ - English#EN-Alto-2#Vibrato#A Thousand Years
359
+ - English#EN-Alto-2#Glissando#A Thousand Years
360
+ - English#EN-Alto-2#Mixed_Voice_and_Falsetto#All of Me
361
+ test_set_name: test
362
+ timesteps: 1000
363
+ train_set_name: train
364
+ train_sets: ''
365
+ use_gt_dur: false
366
+ use_gt_f0: false
367
+ use_nsf: true
368
+ use_ph_postnet: true
369
+ use_pitch_embed: true
370
+ use_pos_embed: true
371
+ use_spk_embed: false
372
+ use_spk_id: true
373
+ use_spk_prompt: false
374
+ use_uv: true
375
+ use_word_input: false
376
+ val_check_interval: 10000
377
+ valid_infer_interval: 10000
378
+ valid_monitor_key: val_loss
379
+ valid_monitor_mode: min
380
+ valid_prefixes:
381
+ - "Chinese#ZH-Alto-1#Breathy#\u4E0D\u518D\u89C1"
382
+ - "Chinese#ZH-Tenor-1#Pharyngeal#\u4E0D\u4E3A\u8C01\u800C\u4F5C\u7684\u6B4C"
383
+ - "Chinese#ZH-Alto-1#Vibrato#\u4E0D\u518D\u89C1"
384
+ - "Chinese#ZH-Tenor-1#Glissando#\u4E0D\u67D3"
385
+ - "Chinese#ZH-Alto-1#Mixed_Voice_and_Falsetto#\u4E00\u6B21\u5C31\u597D"
386
+ - English#EN-Alto-1#Breathy#all is found
387
+ - English#EN-Alto-1#Pharyngeal#beauty and the beast
388
+ - English#EN-Alto-2#Vibrato#A Thousand Years
389
+ - English#EN-Alto-2#Glissando#A Thousand Years
390
+ - English#EN-Alto-2#Mixed_Voice_and_Falsetto#All of Me
391
+ valid_set_name: valid
392
+ vocoder: HifiGAN_NSF
393
+ vocoder_ckpt: checkpoints/hifigan
394
+ warmup_updates: 4000
395
+ weight_decay: 0
396
+ win_size: 1024
397
+ word_dict_size: 10000
398
+ work_dir: checkpoints/stage2
bilingual/stage2/model_ckpt_steps_160000.ckpt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0613bd31b2234d8dca83ae4a48a4e3141931cf328cb21e33cfdba7c4dad8bfc7
3
+ size 288237596