Commit
•
17ee9de
1
Parent(s):
7dbd2dd
Update model
Browse files- README.md +369 -0
- exp/e/tts_train_fastspeech2_raw_phn_none/config.yaml +291 -0
- exp/e/tts_train_fastspeech2_raw_phn_none/images/backward_time.png +0 -0
- exp/e/tts_train_fastspeech2_raw_phn_none/images/decoder_alpha.png +0 -0
- exp/e/tts_train_fastspeech2_raw_phn_none/images/duration_loss.png +0 -0
- exp/e/tts_train_fastspeech2_raw_phn_none/images/encoder_alpha.png +0 -0
- exp/e/tts_train_fastspeech2_raw_phn_none/images/energy_loss.png +0 -0
- exp/e/tts_train_fastspeech2_raw_phn_none/images/forward_time.png +0 -0
- exp/e/tts_train_fastspeech2_raw_phn_none/images/gpu_max_cached_mem_GB.png +0 -0
- exp/e/tts_train_fastspeech2_raw_phn_none/images/iter_time.png +0 -0
- exp/e/tts_train_fastspeech2_raw_phn_none/images/l1_loss.png +0 -0
- exp/e/tts_train_fastspeech2_raw_phn_none/images/loss.png +0 -0
- exp/e/tts_train_fastspeech2_raw_phn_none/images/optim0_lr0.png +0 -0
- exp/e/tts_train_fastspeech2_raw_phn_none/images/optim_step_time.png +0 -0
- exp/e/tts_train_fastspeech2_raw_phn_none/images/pitch_loss.png +0 -0
- exp/e/tts_train_fastspeech2_raw_phn_none/images/train_time.png +0 -0
- exp/e/tts_train_fastspeech2_raw_phn_none/train.loss.ave_5best.pth +3 -0
- exp/e/tts_train_tacotron2_raw_phn_none/decode_use_teacher_forcingtrue_train.loss.ave/stats/train/energy_stats.npz +0 -0
- exp/e/tts_train_tacotron2_raw_phn_none/decode_use_teacher_forcingtrue_train.loss.ave/stats/train/feats_stats.npz +0 -0
- exp/e/tts_train_tacotron2_raw_phn_none/decode_use_teacher_forcingtrue_train.loss.ave/stats/train/pitch_stats.npz +0 -0
- meta.yaml +8 -0
README.md
ADDED
@@ -0,0 +1,369 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
---
|
2 |
+
tags:
|
3 |
+
- espnet
|
4 |
+
- audio
|
5 |
+
- text-to-speech
|
6 |
+
language: en
|
7 |
+
datasets:
|
8 |
+
- talromur
|
9 |
+
license: cc-by-4.0
|
10 |
+
---
|
11 |
+
|
12 |
+
## ESPnet2 TTS model
|
13 |
+
|
14 |
+
### `espnet/GunnarThor_talromur_e_fastspeech2`
|
15 |
+
|
16 |
+
This model was trained by Gunnar Thor using talromur recipe in [espnet](https://github.com/espnet/espnet/).
|
17 |
+
|
18 |
+
### Demo: How to use in ESPnet2
|
19 |
+
|
20 |
+
```bash
|
21 |
+
cd espnet
|
22 |
+
git checkout 49a284e69308d81c142b89795de255b4ce290c54
|
23 |
+
pip install -e .
|
24 |
+
cd egs2/talromur/tts1
|
25 |
+
./run.sh --skip_data_prep false --skip_train true --download_model espnet/GunnarThor_talromur_e_fastspeech2
|
26 |
+
```
|
27 |
+
|
28 |
+
|
29 |
+
|
30 |
+
## TTS config
|
31 |
+
|
32 |
+
<details><summary>expand</summary>
|
33 |
+
|
34 |
+
```
|
35 |
+
config: conf/tuning/train_fastspeech2.yaml
|
36 |
+
print_config: false
|
37 |
+
log_level: INFO
|
38 |
+
dry_run: false
|
39 |
+
iterator_type: sequence
|
40 |
+
output_dir: exp/e/tts_train_fastspeech2_raw_phn_none
|
41 |
+
ngpu: 1
|
42 |
+
seed: 0
|
43 |
+
num_workers: 1
|
44 |
+
num_att_plot: 3
|
45 |
+
dist_backend: nccl
|
46 |
+
dist_init_method: env://
|
47 |
+
dist_world_size: null
|
48 |
+
dist_rank: null
|
49 |
+
local_rank: 0
|
50 |
+
dist_master_addr: null
|
51 |
+
dist_master_port: null
|
52 |
+
dist_launcher: null
|
53 |
+
multiprocessing_distributed: false
|
54 |
+
unused_parameters: false
|
55 |
+
sharded_ddp: false
|
56 |
+
cudnn_enabled: true
|
57 |
+
cudnn_benchmark: false
|
58 |
+
cudnn_deterministic: true
|
59 |
+
collect_stats: false
|
60 |
+
write_collected_feats: false
|
61 |
+
max_epoch: 100
|
62 |
+
patience: null
|
63 |
+
val_scheduler_criterion:
|
64 |
+
- valid
|
65 |
+
- loss
|
66 |
+
early_stopping_criterion:
|
67 |
+
- valid
|
68 |
+
- loss
|
69 |
+
- min
|
70 |
+
best_model_criterion:
|
71 |
+
- - valid
|
72 |
+
- loss
|
73 |
+
- min
|
74 |
+
- - train
|
75 |
+
- loss
|
76 |
+
- min
|
77 |
+
keep_nbest_models: 5
|
78 |
+
nbest_averaging_interval: 0
|
79 |
+
grad_clip: 1.0
|
80 |
+
grad_clip_type: 2.0
|
81 |
+
grad_noise: false
|
82 |
+
accum_grad: 8
|
83 |
+
no_forward_run: false
|
84 |
+
resume: true
|
85 |
+
train_dtype: float32
|
86 |
+
use_amp: false
|
87 |
+
log_interval: null
|
88 |
+
use_matplotlib: true
|
89 |
+
use_tensorboard: true
|
90 |
+
use_wandb: false
|
91 |
+
wandb_project: null
|
92 |
+
wandb_id: null
|
93 |
+
wandb_entity: null
|
94 |
+
wandb_name: null
|
95 |
+
wandb_model_log_interval: -1
|
96 |
+
detect_anomaly: false
|
97 |
+
pretrain_path: null
|
98 |
+
init_param: []
|
99 |
+
ignore_init_mismatch: false
|
100 |
+
freeze_param: []
|
101 |
+
num_iters_per_epoch: 800
|
102 |
+
batch_size: 20
|
103 |
+
valid_batch_size: null
|
104 |
+
batch_bins: 2500000
|
105 |
+
valid_batch_bins: null
|
106 |
+
train_shape_file:
|
107 |
+
- exp/e/tts_train_tacotron2_raw_phn_none/decode_use_teacher_forcingtrue_train.loss.ave/stats/train/text_shape.phn
|
108 |
+
- exp/e/tts_train_tacotron2_raw_phn_none/decode_use_teacher_forcingtrue_train.loss.ave/stats/train/speech_shape
|
109 |
+
valid_shape_file:
|
110 |
+
- exp/e/tts_train_tacotron2_raw_phn_none/decode_use_teacher_forcingtrue_train.loss.ave/stats/valid/text_shape.phn
|
111 |
+
- exp/e/tts_train_tacotron2_raw_phn_none/decode_use_teacher_forcingtrue_train.loss.ave/stats/valid/speech_shape
|
112 |
+
batch_type: numel
|
113 |
+
valid_batch_type: null
|
114 |
+
fold_length:
|
115 |
+
- 150
|
116 |
+
- 204800
|
117 |
+
sort_in_batch: descending
|
118 |
+
sort_batch: descending
|
119 |
+
multiple_iterator: false
|
120 |
+
chunk_length: 500
|
121 |
+
chunk_shift_ratio: 0.5
|
122 |
+
num_cache_chunks: 1024
|
123 |
+
train_data_path_and_name_and_type:
|
124 |
+
- - dump/raw/train_e_phn/text
|
125 |
+
- text
|
126 |
+
- text
|
127 |
+
- - exp/e/tts_train_tacotron2_raw_phn_none/decode_use_teacher_forcingtrue_train.loss.ave/train_e_phn/durations
|
128 |
+
- durations
|
129 |
+
- text_int
|
130 |
+
- - dump/raw/train_e_phn/wav.scp
|
131 |
+
- speech
|
132 |
+
- sound
|
133 |
+
valid_data_path_and_name_and_type:
|
134 |
+
- - dump/raw/dev_e_phn/text
|
135 |
+
- text
|
136 |
+
- text
|
137 |
+
- - exp/e/tts_train_tacotron2_raw_phn_none/decode_use_teacher_forcingtrue_train.loss.ave/dev_e_phn/durations
|
138 |
+
- durations
|
139 |
+
- text_int
|
140 |
+
- - dump/raw/dev_e_phn/wav.scp
|
141 |
+
- speech
|
142 |
+
- sound
|
143 |
+
allow_variable_data_keys: false
|
144 |
+
max_cache_size: 0.0
|
145 |
+
max_cache_fd: 32
|
146 |
+
valid_max_cache_size: null
|
147 |
+
optim: adam
|
148 |
+
optim_conf:
|
149 |
+
lr: 1.0
|
150 |
+
scheduler: noamlr
|
151 |
+
scheduler_conf:
|
152 |
+
model_size: 384
|
153 |
+
warmup_steps: 4000
|
154 |
+
token_list:
|
155 |
+
- <blank>
|
156 |
+
- <unk>
|
157 |
+
- ','
|
158 |
+
- .
|
159 |
+
- r
|
160 |
+
- t
|
161 |
+
- n
|
162 |
+
- a0
|
163 |
+
- s
|
164 |
+
- I0
|
165 |
+
- D
|
166 |
+
- l
|
167 |
+
- Y0
|
168 |
+
- m
|
169 |
+
- v
|
170 |
+
- h
|
171 |
+
- E1
|
172 |
+
- k
|
173 |
+
- a:1
|
174 |
+
- E:1
|
175 |
+
- j
|
176 |
+
- f
|
177 |
+
- T
|
178 |
+
- G
|
179 |
+
- a1
|
180 |
+
- p
|
181 |
+
- c
|
182 |
+
- au:1
|
183 |
+
- i:1
|
184 |
+
- O:1
|
185 |
+
- E0
|
186 |
+
- I:1
|
187 |
+
- r_0
|
188 |
+
- I1
|
189 |
+
- t_h
|
190 |
+
- k_h
|
191 |
+
- Y1
|
192 |
+
- i0
|
193 |
+
- ei1
|
194 |
+
- u:1
|
195 |
+
- ou:1
|
196 |
+
- ei:1
|
197 |
+
- O1
|
198 |
+
- N
|
199 |
+
- l_0
|
200 |
+
- '91'
|
201 |
+
- ou0
|
202 |
+
- ai0
|
203 |
+
- n_0
|
204 |
+
- au1
|
205 |
+
- O0
|
206 |
+
- ou1
|
207 |
+
- ai:1
|
208 |
+
- ei0
|
209 |
+
- '9:1'
|
210 |
+
- ai1
|
211 |
+
- i1
|
212 |
+
- c_h
|
213 |
+
- '90'
|
214 |
+
- au0
|
215 |
+
- x
|
216 |
+
- C
|
217 |
+
- p_h
|
218 |
+
- u0
|
219 |
+
- 9i:1
|
220 |
+
- Y:1
|
221 |
+
- 9i1
|
222 |
+
- J
|
223 |
+
- u1
|
224 |
+
- 9i0
|
225 |
+
- N_0
|
226 |
+
- m_0
|
227 |
+
- J_0
|
228 |
+
- Oi1
|
229 |
+
- Yi0
|
230 |
+
- Yi1
|
231 |
+
- Oi0
|
232 |
+
- '9:0'
|
233 |
+
- au:0
|
234 |
+
- E:0
|
235 |
+
- <sos/eos>
|
236 |
+
odim: null
|
237 |
+
model_conf: {}
|
238 |
+
use_preprocessor: true
|
239 |
+
token_type: phn
|
240 |
+
bpemodel: null
|
241 |
+
non_linguistic_symbols: null
|
242 |
+
cleaner: null
|
243 |
+
g2p: null
|
244 |
+
feats_extract: fbank
|
245 |
+
feats_extract_conf:
|
246 |
+
n_fft: 1024
|
247 |
+
hop_length: 256
|
248 |
+
win_length: null
|
249 |
+
fs: 22050
|
250 |
+
fmin: 80
|
251 |
+
fmax: 7600
|
252 |
+
n_mels: 80
|
253 |
+
normalize: global_mvn
|
254 |
+
normalize_conf:
|
255 |
+
stats_file: exp/e/tts_train_tacotron2_raw_phn_none/decode_use_teacher_forcingtrue_train.loss.ave/stats/train/feats_stats.npz
|
256 |
+
tts: fastspeech2
|
257 |
+
tts_conf:
|
258 |
+
adim: 384
|
259 |
+
aheads: 2
|
260 |
+
elayers: 4
|
261 |
+
eunits: 1536
|
262 |
+
dlayers: 4
|
263 |
+
dunits: 1536
|
264 |
+
positionwise_layer_type: conv1d
|
265 |
+
positionwise_conv_kernel_size: 3
|
266 |
+
duration_predictor_layers: 2
|
267 |
+
duration_predictor_chans: 256
|
268 |
+
duration_predictor_kernel_size: 3
|
269 |
+
postnet_layers: 5
|
270 |
+
postnet_filts: 5
|
271 |
+
postnet_chans: 256
|
272 |
+
use_masking: true
|
273 |
+
use_scaled_pos_enc: true
|
274 |
+
encoder_normalize_before: true
|
275 |
+
decoder_normalize_before: true
|
276 |
+
reduction_factor: 1
|
277 |
+
init_type: xavier_uniform
|
278 |
+
init_enc_alpha: 1.0
|
279 |
+
init_dec_alpha: 1.0
|
280 |
+
transformer_enc_dropout_rate: 0.2
|
281 |
+
transformer_enc_positional_dropout_rate: 0.2
|
282 |
+
transformer_enc_attn_dropout_rate: 0.2
|
283 |
+
transformer_dec_dropout_rate: 0.2
|
284 |
+
transformer_dec_positional_dropout_rate: 0.2
|
285 |
+
transformer_dec_attn_dropout_rate: 0.2
|
286 |
+
pitch_predictor_layers: 5
|
287 |
+
pitch_predictor_chans: 256
|
288 |
+
pitch_predictor_kernel_size: 5
|
289 |
+
pitch_predictor_dropout: 0.5
|
290 |
+
pitch_embed_kernel_size: 1
|
291 |
+
pitch_embed_dropout: 0.0
|
292 |
+
stop_gradient_from_pitch_predictor: true
|
293 |
+
energy_predictor_layers: 2
|
294 |
+
energy_predictor_chans: 256
|
295 |
+
energy_predictor_kernel_size: 3
|
296 |
+
energy_predictor_dropout: 0.5
|
297 |
+
energy_embed_kernel_size: 1
|
298 |
+
energy_embed_dropout: 0.0
|
299 |
+
stop_gradient_from_energy_predictor: false
|
300 |
+
pitch_extract: dio
|
301 |
+
pitch_extract_conf:
|
302 |
+
fs: 22050
|
303 |
+
n_fft: 1024
|
304 |
+
hop_length: 256
|
305 |
+
f0max: 400
|
306 |
+
f0min: 80
|
307 |
+
reduction_factor: 1
|
308 |
+
pitch_normalize: global_mvn
|
309 |
+
pitch_normalize_conf:
|
310 |
+
stats_file: exp/e/tts_train_tacotron2_raw_phn_none/decode_use_teacher_forcingtrue_train.loss.ave/stats/train/pitch_stats.npz
|
311 |
+
energy_extract: energy
|
312 |
+
energy_extract_conf:
|
313 |
+
fs: 22050
|
314 |
+
n_fft: 1024
|
315 |
+
hop_length: 256
|
316 |
+
win_length: null
|
317 |
+
reduction_factor: 1
|
318 |
+
energy_normalize: global_mvn
|
319 |
+
energy_normalize_conf:
|
320 |
+
stats_file: exp/e/tts_train_tacotron2_raw_phn_none/decode_use_teacher_forcingtrue_train.loss.ave/stats/train/energy_stats.npz
|
321 |
+
required:
|
322 |
+
- output_dir
|
323 |
+
- token_list
|
324 |
+
version: 0.10.7a1
|
325 |
+
distributed: false
|
326 |
+
```
|
327 |
+
|
328 |
+
</details>
|
329 |
+
|
330 |
+
|
331 |
+
|
332 |
+
### Citing ESPnet
|
333 |
+
|
334 |
+
```BibTex
|
335 |
+
@inproceedings{watanabe2018espnet,
|
336 |
+
author={Shinji Watanabe and Takaaki Hori and Shigeki Karita and Tomoki Hayashi and Jiro Nishitoba and Yuya Unno and Nelson Yalta and Jahn Heymann and Matthew Wiesner and Nanxin Chen and Adithya Renduchintala and Tsubasa Ochiai},
|
337 |
+
title={{ESPnet}: End-to-End Speech Processing Toolkit},
|
338 |
+
year={2018},
|
339 |
+
booktitle={Proceedings of Interspeech},
|
340 |
+
pages={2207--2211},
|
341 |
+
doi={10.21437/Interspeech.2018-1456},
|
342 |
+
url={http://dx.doi.org/10.21437/Interspeech.2018-1456}
|
343 |
+
}
|
344 |
+
|
345 |
+
|
346 |
+
|
347 |
+
|
348 |
+
@inproceedings{hayashi2020espnet,
|
349 |
+
title={{Espnet-TTS}: Unified, reproducible, and integratable open source end-to-end text-to-speech toolkit},
|
350 |
+
author={Hayashi, Tomoki and Yamamoto, Ryuichi and Inoue, Katsuki and Yoshimura, Takenori and Watanabe, Shinji and Toda, Tomoki and Takeda, Kazuya and Zhang, Yu and Tan, Xu},
|
351 |
+
booktitle={Proceedings of IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)},
|
352 |
+
pages={7654--7658},
|
353 |
+
year={2020},
|
354 |
+
organization={IEEE}
|
355 |
+
}
|
356 |
+
```
|
357 |
+
|
358 |
+
or arXiv:
|
359 |
+
|
360 |
+
```bibtex
|
361 |
+
@misc{watanabe2018espnet,
|
362 |
+
title={ESPnet: End-to-End Speech Processing Toolkit},
|
363 |
+
author={Shinji Watanabe and Takaaki Hori and Shigeki Karita and Tomoki Hayashi and Jiro Nishitoba and Yuya Unno and Nelson Yalta and Jahn Heymann and Matthew Wiesner and Nanxin Chen and Adithya Renduchintala and Tsubasa Ochiai},
|
364 |
+
year={2018},
|
365 |
+
eprint={1804.00015},
|
366 |
+
archivePrefix={arXiv},
|
367 |
+
primaryClass={cs.CL}
|
368 |
+
}
|
369 |
+
```
|
exp/e/tts_train_fastspeech2_raw_phn_none/config.yaml
ADDED
@@ -0,0 +1,291 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
config: conf/tuning/train_fastspeech2.yaml
|
2 |
+
print_config: false
|
3 |
+
log_level: INFO
|
4 |
+
dry_run: false
|
5 |
+
iterator_type: sequence
|
6 |
+
output_dir: exp/e/tts_train_fastspeech2_raw_phn_none
|
7 |
+
ngpu: 1
|
8 |
+
seed: 0
|
9 |
+
num_workers: 1
|
10 |
+
num_att_plot: 3
|
11 |
+
dist_backend: nccl
|
12 |
+
dist_init_method: env://
|
13 |
+
dist_world_size: null
|
14 |
+
dist_rank: null
|
15 |
+
local_rank: 0
|
16 |
+
dist_master_addr: null
|
17 |
+
dist_master_port: null
|
18 |
+
dist_launcher: null
|
19 |
+
multiprocessing_distributed: false
|
20 |
+
unused_parameters: false
|
21 |
+
sharded_ddp: false
|
22 |
+
cudnn_enabled: true
|
23 |
+
cudnn_benchmark: false
|
24 |
+
cudnn_deterministic: true
|
25 |
+
collect_stats: false
|
26 |
+
write_collected_feats: false
|
27 |
+
max_epoch: 100
|
28 |
+
patience: null
|
29 |
+
val_scheduler_criterion:
|
30 |
+
- valid
|
31 |
+
- loss
|
32 |
+
early_stopping_criterion:
|
33 |
+
- valid
|
34 |
+
- loss
|
35 |
+
- min
|
36 |
+
best_model_criterion:
|
37 |
+
- - valid
|
38 |
+
- loss
|
39 |
+
- min
|
40 |
+
- - train
|
41 |
+
- loss
|
42 |
+
- min
|
43 |
+
keep_nbest_models: 5
|
44 |
+
nbest_averaging_interval: 0
|
45 |
+
grad_clip: 1.0
|
46 |
+
grad_clip_type: 2.0
|
47 |
+
grad_noise: false
|
48 |
+
accum_grad: 8
|
49 |
+
no_forward_run: false
|
50 |
+
resume: true
|
51 |
+
train_dtype: float32
|
52 |
+
use_amp: false
|
53 |
+
log_interval: null
|
54 |
+
use_matplotlib: true
|
55 |
+
use_tensorboard: true
|
56 |
+
use_wandb: false
|
57 |
+
wandb_project: null
|
58 |
+
wandb_id: null
|
59 |
+
wandb_entity: null
|
60 |
+
wandb_name: null
|
61 |
+
wandb_model_log_interval: -1
|
62 |
+
detect_anomaly: false
|
63 |
+
pretrain_path: null
|
64 |
+
init_param: []
|
65 |
+
ignore_init_mismatch: false
|
66 |
+
freeze_param: []
|
67 |
+
num_iters_per_epoch: 800
|
68 |
+
batch_size: 20
|
69 |
+
valid_batch_size: null
|
70 |
+
batch_bins: 2500000
|
71 |
+
valid_batch_bins: null
|
72 |
+
train_shape_file:
|
73 |
+
- exp/e/tts_train_tacotron2_raw_phn_none/decode_use_teacher_forcingtrue_train.loss.ave/stats/train/text_shape.phn
|
74 |
+
- exp/e/tts_train_tacotron2_raw_phn_none/decode_use_teacher_forcingtrue_train.loss.ave/stats/train/speech_shape
|
75 |
+
valid_shape_file:
|
76 |
+
- exp/e/tts_train_tacotron2_raw_phn_none/decode_use_teacher_forcingtrue_train.loss.ave/stats/valid/text_shape.phn
|
77 |
+
- exp/e/tts_train_tacotron2_raw_phn_none/decode_use_teacher_forcingtrue_train.loss.ave/stats/valid/speech_shape
|
78 |
+
batch_type: numel
|
79 |
+
valid_batch_type: null
|
80 |
+
fold_length:
|
81 |
+
- 150
|
82 |
+
- 204800
|
83 |
+
sort_in_batch: descending
|
84 |
+
sort_batch: descending
|
85 |
+
multiple_iterator: false
|
86 |
+
chunk_length: 500
|
87 |
+
chunk_shift_ratio: 0.5
|
88 |
+
num_cache_chunks: 1024
|
89 |
+
train_data_path_and_name_and_type:
|
90 |
+
- - dump/raw/train_e_phn/text
|
91 |
+
- text
|
92 |
+
- text
|
93 |
+
- - exp/e/tts_train_tacotron2_raw_phn_none/decode_use_teacher_forcingtrue_train.loss.ave/train_e_phn/durations
|
94 |
+
- durations
|
95 |
+
- text_int
|
96 |
+
- - dump/raw/train_e_phn/wav.scp
|
97 |
+
- speech
|
98 |
+
- sound
|
99 |
+
valid_data_path_and_name_and_type:
|
100 |
+
- - dump/raw/dev_e_phn/text
|
101 |
+
- text
|
102 |
+
- text
|
103 |
+
- - exp/e/tts_train_tacotron2_raw_phn_none/decode_use_teacher_forcingtrue_train.loss.ave/dev_e_phn/durations
|
104 |
+
- durations
|
105 |
+
- text_int
|
106 |
+
- - dump/raw/dev_e_phn/wav.scp
|
107 |
+
- speech
|
108 |
+
- sound
|
109 |
+
allow_variable_data_keys: false
|
110 |
+
max_cache_size: 0.0
|
111 |
+
max_cache_fd: 32
|
112 |
+
valid_max_cache_size: null
|
113 |
+
optim: adam
|
114 |
+
optim_conf:
|
115 |
+
lr: 1.0
|
116 |
+
scheduler: noamlr
|
117 |
+
scheduler_conf:
|
118 |
+
model_size: 384
|
119 |
+
warmup_steps: 4000
|
120 |
+
token_list:
|
121 |
+
- <blank>
|
122 |
+
- <unk>
|
123 |
+
- ','
|
124 |
+
- .
|
125 |
+
- r
|
126 |
+
- t
|
127 |
+
- n
|
128 |
+
- a0
|
129 |
+
- s
|
130 |
+
- I0
|
131 |
+
- D
|
132 |
+
- l
|
133 |
+
- Y0
|
134 |
+
- m
|
135 |
+
- v
|
136 |
+
- h
|
137 |
+
- E1
|
138 |
+
- k
|
139 |
+
- a:1
|
140 |
+
- E:1
|
141 |
+
- j
|
142 |
+
- f
|
143 |
+
- T
|
144 |
+
- G
|
145 |
+
- a1
|
146 |
+
- p
|
147 |
+
- c
|
148 |
+
- au:1
|
149 |
+
- i:1
|
150 |
+
- O:1
|
151 |
+
- E0
|
152 |
+
- I:1
|
153 |
+
- r_0
|
154 |
+
- I1
|
155 |
+
- t_h
|
156 |
+
- k_h
|
157 |
+
- Y1
|
158 |
+
- i0
|
159 |
+
- ei1
|
160 |
+
- u:1
|
161 |
+
- ou:1
|
162 |
+
- ei:1
|
163 |
+
- O1
|
164 |
+
- N
|
165 |
+
- l_0
|
166 |
+
- '91'
|
167 |
+
- ou0
|
168 |
+
- ai0
|
169 |
+
- n_0
|
170 |
+
- au1
|
171 |
+
- O0
|
172 |
+
- ou1
|
173 |
+
- ai:1
|
174 |
+
- ei0
|
175 |
+
- '9:1'
|
176 |
+
- ai1
|
177 |
+
- i1
|
178 |
+
- c_h
|
179 |
+
- '90'
|
180 |
+
- au0
|
181 |
+
- x
|
182 |
+
- C
|
183 |
+
- p_h
|
184 |
+
- u0
|
185 |
+
- 9i:1
|
186 |
+
- Y:1
|
187 |
+
- 9i1
|
188 |
+
- J
|
189 |
+
- u1
|
190 |
+
- 9i0
|
191 |
+
- N_0
|
192 |
+
- m_0
|
193 |
+
- J_0
|
194 |
+
- Oi1
|
195 |
+
- Yi0
|
196 |
+
- Yi1
|
197 |
+
- Oi0
|
198 |
+
- '9:0'
|
199 |
+
- au:0
|
200 |
+
- E:0
|
201 |
+
- <sos/eos>
|
202 |
+
odim: null
|
203 |
+
model_conf: {}
|
204 |
+
use_preprocessor: true
|
205 |
+
token_type: phn
|
206 |
+
bpemodel: null
|
207 |
+
non_linguistic_symbols: null
|
208 |
+
cleaner: null
|
209 |
+
g2p: null
|
210 |
+
feats_extract: fbank
|
211 |
+
feats_extract_conf:
|
212 |
+
n_fft: 1024
|
213 |
+
hop_length: 256
|
214 |
+
win_length: null
|
215 |
+
fs: 22050
|
216 |
+
fmin: 80
|
217 |
+
fmax: 7600
|
218 |
+
n_mels: 80
|
219 |
+
normalize: global_mvn
|
220 |
+
normalize_conf:
|
221 |
+
stats_file: exp/e/tts_train_tacotron2_raw_phn_none/decode_use_teacher_forcingtrue_train.loss.ave/stats/train/feats_stats.npz
|
222 |
+
tts: fastspeech2
|
223 |
+
tts_conf:
|
224 |
+
adim: 384
|
225 |
+
aheads: 2
|
226 |
+
elayers: 4
|
227 |
+
eunits: 1536
|
228 |
+
dlayers: 4
|
229 |
+
dunits: 1536
|
230 |
+
positionwise_layer_type: conv1d
|
231 |
+
positionwise_conv_kernel_size: 3
|
232 |
+
duration_predictor_layers: 2
|
233 |
+
duration_predictor_chans: 256
|
234 |
+
duration_predictor_kernel_size: 3
|
235 |
+
postnet_layers: 5
|
236 |
+
postnet_filts: 5
|
237 |
+
postnet_chans: 256
|
238 |
+
use_masking: true
|
239 |
+
use_scaled_pos_enc: true
|
240 |
+
encoder_normalize_before: true
|
241 |
+
decoder_normalize_before: true
|
242 |
+
reduction_factor: 1
|
243 |
+
init_type: xavier_uniform
|
244 |
+
init_enc_alpha: 1.0
|
245 |
+
init_dec_alpha: 1.0
|
246 |
+
transformer_enc_dropout_rate: 0.2
|
247 |
+
transformer_enc_positional_dropout_rate: 0.2
|
248 |
+
transformer_enc_attn_dropout_rate: 0.2
|
249 |
+
transformer_dec_dropout_rate: 0.2
|
250 |
+
transformer_dec_positional_dropout_rate: 0.2
|
251 |
+
transformer_dec_attn_dropout_rate: 0.2
|
252 |
+
pitch_predictor_layers: 5
|
253 |
+
pitch_predictor_chans: 256
|
254 |
+
pitch_predictor_kernel_size: 5
|
255 |
+
pitch_predictor_dropout: 0.5
|
256 |
+
pitch_embed_kernel_size: 1
|
257 |
+
pitch_embed_dropout: 0.0
|
258 |
+
stop_gradient_from_pitch_predictor: true
|
259 |
+
energy_predictor_layers: 2
|
260 |
+
energy_predictor_chans: 256
|
261 |
+
energy_predictor_kernel_size: 3
|
262 |
+
energy_predictor_dropout: 0.5
|
263 |
+
energy_embed_kernel_size: 1
|
264 |
+
energy_embed_dropout: 0.0
|
265 |
+
stop_gradient_from_energy_predictor: false
|
266 |
+
pitch_extract: dio
|
267 |
+
pitch_extract_conf:
|
268 |
+
fs: 22050
|
269 |
+
n_fft: 1024
|
270 |
+
hop_length: 256
|
271 |
+
f0max: 400
|
272 |
+
f0min: 80
|
273 |
+
reduction_factor: 1
|
274 |
+
pitch_normalize: global_mvn
|
275 |
+
pitch_normalize_conf:
|
276 |
+
stats_file: exp/e/tts_train_tacotron2_raw_phn_none/decode_use_teacher_forcingtrue_train.loss.ave/stats/train/pitch_stats.npz
|
277 |
+
energy_extract: energy
|
278 |
+
energy_extract_conf:
|
279 |
+
fs: 22050
|
280 |
+
n_fft: 1024
|
281 |
+
hop_length: 256
|
282 |
+
win_length: null
|
283 |
+
reduction_factor: 1
|
284 |
+
energy_normalize: global_mvn
|
285 |
+
energy_normalize_conf:
|
286 |
+
stats_file: exp/e/tts_train_tacotron2_raw_phn_none/decode_use_teacher_forcingtrue_train.loss.ave/stats/train/energy_stats.npz
|
287 |
+
required:
|
288 |
+
- output_dir
|
289 |
+
- token_list
|
290 |
+
version: 0.10.7a1
|
291 |
+
distributed: false
|
exp/e/tts_train_fastspeech2_raw_phn_none/images/backward_time.png
ADDED
![]() |
exp/e/tts_train_fastspeech2_raw_phn_none/images/decoder_alpha.png
ADDED
![]() |
exp/e/tts_train_fastspeech2_raw_phn_none/images/duration_loss.png
ADDED
![]() |
exp/e/tts_train_fastspeech2_raw_phn_none/images/encoder_alpha.png
ADDED
![]() |
exp/e/tts_train_fastspeech2_raw_phn_none/images/energy_loss.png
ADDED
![]() |
exp/e/tts_train_fastspeech2_raw_phn_none/images/forward_time.png
ADDED
![]() |
exp/e/tts_train_fastspeech2_raw_phn_none/images/gpu_max_cached_mem_GB.png
ADDED
![]() |
exp/e/tts_train_fastspeech2_raw_phn_none/images/iter_time.png
ADDED
![]() |
exp/e/tts_train_fastspeech2_raw_phn_none/images/l1_loss.png
ADDED
![]() |
exp/e/tts_train_fastspeech2_raw_phn_none/images/loss.png
ADDED
![]() |
exp/e/tts_train_fastspeech2_raw_phn_none/images/optim0_lr0.png
ADDED
![]() |
exp/e/tts_train_fastspeech2_raw_phn_none/images/optim_step_time.png
ADDED
![]() |
exp/e/tts_train_fastspeech2_raw_phn_none/images/pitch_loss.png
ADDED
![]() |
exp/e/tts_train_fastspeech2_raw_phn_none/images/train_time.png
ADDED
![]() |
exp/e/tts_train_fastspeech2_raw_phn_none/train.loss.ave_5best.pth
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:0764cc2334b9bc3c3b08539e0af70edf2f5ae247c14006a189593e0d6bf2064d
|
3 |
+
size 148726473
|
exp/e/tts_train_tacotron2_raw_phn_none/decode_use_teacher_forcingtrue_train.loss.ave/stats/train/energy_stats.npz
ADDED
Binary file (770 Bytes). View file
|
|
exp/e/tts_train_tacotron2_raw_phn_none/decode_use_teacher_forcingtrue_train.loss.ave/stats/train/feats_stats.npz
ADDED
Binary file (1.4 kB). View file
|
|
exp/e/tts_train_tacotron2_raw_phn_none/decode_use_teacher_forcingtrue_train.loss.ave/stats/train/pitch_stats.npz
ADDED
Binary file (770 Bytes). View file
|
|
meta.yaml
ADDED
@@ -0,0 +1,8 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
espnet: 0.10.7a1
|
2 |
+
files:
|
3 |
+
model_file: exp/e/tts_train_fastspeech2_raw_phn_none/train.loss.ave_5best.pth
|
4 |
+
python: "3.8.12 (default, Oct 12 2021, 13:49:34) \n[GCC 7.5.0]"
|
5 |
+
timestamp: 1649582828.038759
|
6 |
+
torch: 1.10.2+cu102
|
7 |
+
yaml_files:
|
8 |
+
train_config: exp/e/tts_train_fastspeech2_raw_phn_none/config.yaml
|