ftshijt
commited on
Commit
•
33b6bba
1
Parent(s):
39a662e
Update model
Browse files- README.md +330 -1
- exp/svs_stats_raw_phn_pyopenjtalk_jp/train/feats_stats.npz +3 -0
- exp/svs_stats_raw_phn_pyopenjtalk_jp/train/pitch_stats.npz +3 -0
- exp/svs_train_naive_rnn_dp_raw_phn_pyopenjtalk_jp/263epoch.pth +3 -0
- exp/svs_train_naive_rnn_dp_raw_phn_pyopenjtalk_jp/config.yaml +249 -0
- exp/svs_train_naive_rnn_dp_raw_phn_pyopenjtalk_jp/images/backward_time.png +0 -0
- exp/svs_train_naive_rnn_dp_raw_phn_pyopenjtalk_jp/images/clip.png +0 -0
- exp/svs_train_naive_rnn_dp_raw_phn_pyopenjtalk_jp/images/duration_loss.png +0 -0
- exp/svs_train_naive_rnn_dp_raw_phn_pyopenjtalk_jp/images/forward_time.png +0 -0
- exp/svs_train_naive_rnn_dp_raw_phn_pyopenjtalk_jp/images/gpu_max_cached_mem_GB.png +0 -0
- exp/svs_train_naive_rnn_dp_raw_phn_pyopenjtalk_jp/images/grad_norm.png +0 -0
- exp/svs_train_naive_rnn_dp_raw_phn_pyopenjtalk_jp/images/iter_time.png +0 -0
- exp/svs_train_naive_rnn_dp_raw_phn_pyopenjtalk_jp/images/l1_loss.png +0 -0
- exp/svs_train_naive_rnn_dp_raw_phn_pyopenjtalk_jp/images/loss.png +0 -0
- exp/svs_train_naive_rnn_dp_raw_phn_pyopenjtalk_jp/images/loss_scale.png +0 -0
- exp/svs_train_naive_rnn_dp_raw_phn_pyopenjtalk_jp/images/optim0_lr0.png +0 -0
- exp/svs_train_naive_rnn_dp_raw_phn_pyopenjtalk_jp/images/optim_step_time.png +0 -0
- exp/svs_train_naive_rnn_dp_raw_phn_pyopenjtalk_jp/images/train_time.png +0 -0
- meta.yaml +8 -0
README.md
CHANGED
@@ -1,3 +1,332 @@
|
|
1 |
---
|
2 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
3 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
---
|
2 |
+
tags:
|
3 |
+
- espnet
|
4 |
+
- audio
|
5 |
+
- singing-voice-synthesis
|
6 |
+
language: jp
|
7 |
+
datasets:
|
8 |
+
- kiritan
|
9 |
+
license: cc-by-4.0
|
10 |
---
|
11 |
+
|
12 |
+
## ESPnet2 SVS model
|
13 |
+
|
14 |
+
### `espnet/kiritan_svs_rnn`
|
15 |
+
|
16 |
+
This model was trained by ftshijt using kiritan recipe in [espnet](https://github.com/espnet/espnet/).
|
17 |
+
|
18 |
+
### Demo: How to use in ESPnet2
|
19 |
+
|
20 |
+
Follow the [ESPnet installation instructions](https://espnet.github.io/espnet/installation.html)
|
21 |
+
if you haven't done that already.
|
22 |
+
|
23 |
+
```bash
|
24 |
+
cd espnet
|
25 |
+
git checkout 5c4d7cf7feba8461de2e1080bf82182f0efaef38
|
26 |
+
pip install -e .
|
27 |
+
cd egs2/kiritan/svs1
|
28 |
+
./run.sh --skip_data_prep false --skip_train true --download_model espnet/kiritan_svs_rnn
|
29 |
+
```
|
30 |
+
|
31 |
+
|
32 |
+
|
33 |
+
## SVS config
|
34 |
+
|
35 |
+
<details><summary>expand</summary>
|
36 |
+
|
37 |
+
```
|
38 |
+
config: conf/tuning/train_naive_rnn_dp.yaml
|
39 |
+
print_config: false
|
40 |
+
log_level: INFO
|
41 |
+
drop_last_iter: false
|
42 |
+
dry_run: false
|
43 |
+
iterator_type: sequence
|
44 |
+
valid_iterator_type: null
|
45 |
+
output_dir: exp/svs_train_naive_rnn_dp_raw_phn_pyopenjtalk_jp
|
46 |
+
ngpu: 1
|
47 |
+
seed: 0
|
48 |
+
num_workers: 8
|
49 |
+
num_att_plot: 3
|
50 |
+
dist_backend: nccl
|
51 |
+
dist_init_method: env://
|
52 |
+
dist_world_size: null
|
53 |
+
dist_rank: null
|
54 |
+
local_rank: 0
|
55 |
+
dist_master_addr: null
|
56 |
+
dist_master_port: null
|
57 |
+
dist_launcher: null
|
58 |
+
multiprocessing_distributed: false
|
59 |
+
unused_parameters: false
|
60 |
+
sharded_ddp: false
|
61 |
+
cudnn_enabled: true
|
62 |
+
cudnn_benchmark: false
|
63 |
+
cudnn_deterministic: true
|
64 |
+
collect_stats: false
|
65 |
+
write_collected_feats: false
|
66 |
+
max_epoch: 500
|
67 |
+
patience: null
|
68 |
+
val_scheduler_criterion:
|
69 |
+
- valid
|
70 |
+
- loss
|
71 |
+
early_stopping_criterion:
|
72 |
+
- valid
|
73 |
+
- loss
|
74 |
+
- min
|
75 |
+
best_model_criterion:
|
76 |
+
- - valid
|
77 |
+
- loss
|
78 |
+
- min
|
79 |
+
- - train
|
80 |
+
- loss
|
81 |
+
- min
|
82 |
+
keep_nbest_models: 2
|
83 |
+
nbest_averaging_interval: 0
|
84 |
+
grad_clip: 1.0
|
85 |
+
grad_clip_type: 2.0
|
86 |
+
grad_noise: false
|
87 |
+
accum_grad: 1
|
88 |
+
no_forward_run: false
|
89 |
+
resume: true
|
90 |
+
train_dtype: float32
|
91 |
+
use_amp: false
|
92 |
+
log_interval: null
|
93 |
+
use_matplotlib: true
|
94 |
+
use_tensorboard: true
|
95 |
+
create_graph_in_tensorboard: false
|
96 |
+
use_wandb: false
|
97 |
+
wandb_project: null
|
98 |
+
wandb_id: null
|
99 |
+
wandb_entity: null
|
100 |
+
wandb_name: null
|
101 |
+
wandb_model_log_interval: -1
|
102 |
+
detect_anomaly: false
|
103 |
+
use_lora: false
|
104 |
+
save_lora_only: true
|
105 |
+
lora_conf: {}
|
106 |
+
pretrain_path: null
|
107 |
+
init_param: []
|
108 |
+
ignore_init_mismatch: false
|
109 |
+
freeze_param: []
|
110 |
+
num_iters_per_epoch: null
|
111 |
+
batch_size: 16
|
112 |
+
valid_batch_size: null
|
113 |
+
batch_bins: 1000000
|
114 |
+
valid_batch_bins: null
|
115 |
+
train_shape_file:
|
116 |
+
- exp/svs_stats_raw_phn_pyopenjtalk_jp/train/text_shape.phn
|
117 |
+
- exp/svs_stats_raw_phn_pyopenjtalk_jp/train/singing_shape
|
118 |
+
valid_shape_file:
|
119 |
+
- exp/svs_stats_raw_phn_pyopenjtalk_jp/valid/text_shape.phn
|
120 |
+
- exp/svs_stats_raw_phn_pyopenjtalk_jp/valid/singing_shape
|
121 |
+
batch_type: sorted
|
122 |
+
valid_batch_type: null
|
123 |
+
fold_length:
|
124 |
+
- 150
|
125 |
+
- 240000
|
126 |
+
sort_in_batch: descending
|
127 |
+
shuffle_within_batch: false
|
128 |
+
sort_batch: descending
|
129 |
+
multiple_iterator: false
|
130 |
+
chunk_length: 500
|
131 |
+
chunk_shift_ratio: 0.5
|
132 |
+
num_cache_chunks: 1024
|
133 |
+
chunk_excluded_key_prefixes: []
|
134 |
+
chunk_default_fs: null
|
135 |
+
train_data_path_and_name_and_type:
|
136 |
+
- - dump/raw/tr_no_dev/text
|
137 |
+
- text
|
138 |
+
- text
|
139 |
+
- - dump/raw/tr_no_dev/wav.scp
|
140 |
+
- singing
|
141 |
+
- sound
|
142 |
+
- - dump/raw/tr_no_dev/label
|
143 |
+
- label
|
144 |
+
- duration
|
145 |
+
- - dump/raw/tr_no_dev/score.scp
|
146 |
+
- score
|
147 |
+
- score
|
148 |
+
valid_data_path_and_name_and_type:
|
149 |
+
- - dump/raw/dev/text
|
150 |
+
- text
|
151 |
+
- text
|
152 |
+
- - dump/raw/dev/wav.scp
|
153 |
+
- singing
|
154 |
+
- sound
|
155 |
+
- - dump/raw/dev/label
|
156 |
+
- label
|
157 |
+
- duration
|
158 |
+
- - dump/raw/dev/score.scp
|
159 |
+
- score
|
160 |
+
- score
|
161 |
+
allow_variable_data_keys: false
|
162 |
+
max_cache_size: 0.0
|
163 |
+
max_cache_fd: 32
|
164 |
+
allow_multi_rates: false
|
165 |
+
valid_max_cache_size: null
|
166 |
+
exclude_weight_decay: false
|
167 |
+
exclude_weight_decay_conf: {}
|
168 |
+
optim: adam
|
169 |
+
optim_conf:
|
170 |
+
lr: 0.001
|
171 |
+
eps: 1.0e-06
|
172 |
+
weight_decay: 0.0
|
173 |
+
scheduler: null
|
174 |
+
scheduler_conf: {}
|
175 |
+
token_list:
|
176 |
+
- <blank>
|
177 |
+
- <unk>
|
178 |
+
- pau
|
179 |
+
- a
|
180 |
+
- i
|
181 |
+
- o
|
182 |
+
- e
|
183 |
+
- u
|
184 |
+
- k
|
185 |
+
- n
|
186 |
+
- r
|
187 |
+
- t
|
188 |
+
- m
|
189 |
+
- d
|
190 |
+
- s
|
191 |
+
- N
|
192 |
+
- sh
|
193 |
+
- g
|
194 |
+
- y
|
195 |
+
- b
|
196 |
+
- w
|
197 |
+
- cl
|
198 |
+
- ts
|
199 |
+
- z
|
200 |
+
- ch
|
201 |
+
- j
|
202 |
+
- h
|
203 |
+
- f
|
204 |
+
- p
|
205 |
+
- ky
|
206 |
+
- ry
|
207 |
+
- hy
|
208 |
+
- py
|
209 |
+
- ny
|
210 |
+
- <sos/eos>
|
211 |
+
odim: null
|
212 |
+
model_conf: {}
|
213 |
+
use_preprocessor: true
|
214 |
+
token_type: phn
|
215 |
+
bpemodel: null
|
216 |
+
non_linguistic_symbols: null
|
217 |
+
cleaner: null
|
218 |
+
g2p: pyopenjtalk
|
219 |
+
fs: 24000
|
220 |
+
score_feats_extract: syllable_score_feats
|
221 |
+
score_feats_extract_conf:
|
222 |
+
fs: 24000
|
223 |
+
n_fft: 2048
|
224 |
+
win_length: 1200
|
225 |
+
hop_length: 300
|
226 |
+
feats_extract: fbank
|
227 |
+
feats_extract_conf:
|
228 |
+
n_fft: 2048
|
229 |
+
hop_length: 300
|
230 |
+
win_length: 1200
|
231 |
+
fs: 24000
|
232 |
+
fmin: 80
|
233 |
+
fmax: 7600
|
234 |
+
n_mels: 80
|
235 |
+
normalize: global_mvn
|
236 |
+
normalize_conf:
|
237 |
+
stats_file: exp/svs_stats_raw_phn_pyopenjtalk_jp/train/feats_stats.npz
|
238 |
+
svs: naive_rnn_dp
|
239 |
+
svs_conf:
|
240 |
+
midi_dim: 129
|
241 |
+
embed_dim: 512
|
242 |
+
duration_dim: 500
|
243 |
+
eprenet_conv_layers: 0
|
244 |
+
eprenet_conv_chans: 256
|
245 |
+
eprenet_conv_filts: 3
|
246 |
+
elayers: 3
|
247 |
+
eunits: 256
|
248 |
+
ebidirectional: true
|
249 |
+
midi_embed_integration_type: add
|
250 |
+
dlayers: 2
|
251 |
+
dunits: 256
|
252 |
+
dbidirectional: true
|
253 |
+
postnet_layers: 5
|
254 |
+
postnet_chans: 512
|
255 |
+
postnet_filts: 5
|
256 |
+
use_batch_norm: true
|
257 |
+
reduction_factor: 1
|
258 |
+
eprenet_dropout_rate: 0.2
|
259 |
+
edropout_rate: 0.1
|
260 |
+
ddropout_rate: 0.1
|
261 |
+
postnet_dropout_rate: 0.5
|
262 |
+
init_type: pytorch
|
263 |
+
use_masking: true
|
264 |
+
pitch_extract: dio
|
265 |
+
pitch_extract_conf:
|
266 |
+
use_token_averaged_f0: false
|
267 |
+
fs: 24000
|
268 |
+
n_fft: 2048
|
269 |
+
hop_length: 300
|
270 |
+
f0max: 800
|
271 |
+
f0min: 80
|
272 |
+
reduction_factor: 1
|
273 |
+
pitch_normalize: global_mvn
|
274 |
+
pitch_normalize_conf:
|
275 |
+
stats_file: exp/svs_stats_raw_phn_pyopenjtalk_jp/train/pitch_stats.npz
|
276 |
+
ying_extract: null
|
277 |
+
ying_extract_conf: {}
|
278 |
+
energy_extract: null
|
279 |
+
energy_extract_conf: {}
|
280 |
+
energy_normalize: null
|
281 |
+
energy_normalize_conf: {}
|
282 |
+
required:
|
283 |
+
- output_dir
|
284 |
+
- token_list
|
285 |
+
version: '202310'
|
286 |
+
distributed: false
|
287 |
+
```
|
288 |
+
|
289 |
+
</details>
|
290 |
+
|
291 |
+
|
292 |
+
|
293 |
+
### Citing ESPnet
|
294 |
+
|
295 |
+
```BibTex
|
296 |
+
@inproceedings{watanabe2018espnet,
|
297 |
+
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},
|
298 |
+
title={{ESPnet}: End-to-End Speech Processing Toolkit},
|
299 |
+
year={2018},
|
300 |
+
booktitle={Proceedings of Interspeech},
|
301 |
+
pages={2207--2211},
|
302 |
+
doi={10.21437/Interspeech.2018-1456},
|
303 |
+
url={http://dx.doi.org/10.21437/Interspeech.2018-1456}
|
304 |
+
}
|
305 |
+
|
306 |
+
|
307 |
+
|
308 |
+
|
309 |
+
|
310 |
+
|
311 |
+
@inproceedings{shi22d_interspeech,
|
312 |
+
author={Jiatong Shi and Shuai Guo and Tao Qian and Tomoki Hayashi and Yuning Wu and Fangzheng Xu and Xuankai Chang and Huazhe Li and Peter Wu and Shinji Watanabe and Qin Jin},
|
313 |
+
title={{Muskits: an End-to-end Music Processing Toolkit for Singing Voice Synthesis}},
|
314 |
+
year=2022,
|
315 |
+
booktitle={Proc. Interspeech 2022},
|
316 |
+
pages={4277--4281},
|
317 |
+
doi={10.21437/Interspeech.2022-10039}
|
318 |
+
}
|
319 |
+
```
|
320 |
+
|
321 |
+
or arXiv:
|
322 |
+
|
323 |
+
```bibtex
|
324 |
+
@misc{watanabe2018espnet,
|
325 |
+
title={ESPnet: End-to-End Speech Processing Toolkit},
|
326 |
+
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},
|
327 |
+
year={2018},
|
328 |
+
eprint={1804.00015},
|
329 |
+
archivePrefix={arXiv},
|
330 |
+
primaryClass={cs.CL}
|
331 |
+
}
|
332 |
+
```
|
exp/svs_stats_raw_phn_pyopenjtalk_jp/train/feats_stats.npz
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:b3f8760a03a38e9f6aedceafc940853e135de03be88dc0f400f80111012ae2f4
|
3 |
+
size 1402
|
exp/svs_stats_raw_phn_pyopenjtalk_jp/train/pitch_stats.npz
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:472f44554816456248d361dd80b0a2a3d17c6dc420486a72fd7a0eedb2144f99
|
3 |
+
size 770
|
exp/svs_train_naive_rnn_dp_raw_phn_pyopenjtalk_jp/263epoch.pth
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:ca72943a6a697d70db0d30cc6dcea6418c3d56857ca6c602eb5a177fe5b96a9a
|
3 |
+
size 86890167
|
exp/svs_train_naive_rnn_dp_raw_phn_pyopenjtalk_jp/config.yaml
ADDED
@@ -0,0 +1,249 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
config: conf/tuning/train_naive_rnn_dp.yaml
|
2 |
+
print_config: false
|
3 |
+
log_level: INFO
|
4 |
+
drop_last_iter: false
|
5 |
+
dry_run: false
|
6 |
+
iterator_type: sequence
|
7 |
+
valid_iterator_type: null
|
8 |
+
output_dir: exp/svs_train_naive_rnn_dp_raw_phn_pyopenjtalk_jp
|
9 |
+
ngpu: 1
|
10 |
+
seed: 0
|
11 |
+
num_workers: 8
|
12 |
+
num_att_plot: 3
|
13 |
+
dist_backend: nccl
|
14 |
+
dist_init_method: env://
|
15 |
+
dist_world_size: null
|
16 |
+
dist_rank: null
|
17 |
+
local_rank: 0
|
18 |
+
dist_master_addr: null
|
19 |
+
dist_master_port: null
|
20 |
+
dist_launcher: null
|
21 |
+
multiprocessing_distributed: false
|
22 |
+
unused_parameters: false
|
23 |
+
sharded_ddp: false
|
24 |
+
cudnn_enabled: true
|
25 |
+
cudnn_benchmark: false
|
26 |
+
cudnn_deterministic: true
|
27 |
+
collect_stats: false
|
28 |
+
write_collected_feats: false
|
29 |
+
max_epoch: 500
|
30 |
+
patience: null
|
31 |
+
val_scheduler_criterion:
|
32 |
+
- valid
|
33 |
+
- loss
|
34 |
+
early_stopping_criterion:
|
35 |
+
- valid
|
36 |
+
- loss
|
37 |
+
- min
|
38 |
+
best_model_criterion:
|
39 |
+
- - valid
|
40 |
+
- loss
|
41 |
+
- min
|
42 |
+
- - train
|
43 |
+
- loss
|
44 |
+
- min
|
45 |
+
keep_nbest_models: 2
|
46 |
+
nbest_averaging_interval: 0
|
47 |
+
grad_clip: 1.0
|
48 |
+
grad_clip_type: 2.0
|
49 |
+
grad_noise: false
|
50 |
+
accum_grad: 1
|
51 |
+
no_forward_run: false
|
52 |
+
resume: true
|
53 |
+
train_dtype: float32
|
54 |
+
use_amp: false
|
55 |
+
log_interval: null
|
56 |
+
use_matplotlib: true
|
57 |
+
use_tensorboard: true
|
58 |
+
create_graph_in_tensorboard: false
|
59 |
+
use_wandb: false
|
60 |
+
wandb_project: null
|
61 |
+
wandb_id: null
|
62 |
+
wandb_entity: null
|
63 |
+
wandb_name: null
|
64 |
+
wandb_model_log_interval: -1
|
65 |
+
detect_anomaly: false
|
66 |
+
use_lora: false
|
67 |
+
save_lora_only: true
|
68 |
+
lora_conf: {}
|
69 |
+
pretrain_path: null
|
70 |
+
init_param: []
|
71 |
+
ignore_init_mismatch: false
|
72 |
+
freeze_param: []
|
73 |
+
num_iters_per_epoch: null
|
74 |
+
batch_size: 16
|
75 |
+
valid_batch_size: null
|
76 |
+
batch_bins: 1000000
|
77 |
+
valid_batch_bins: null
|
78 |
+
train_shape_file:
|
79 |
+
- exp/svs_stats_raw_phn_pyopenjtalk_jp/train/text_shape.phn
|
80 |
+
- exp/svs_stats_raw_phn_pyopenjtalk_jp/train/singing_shape
|
81 |
+
valid_shape_file:
|
82 |
+
- exp/svs_stats_raw_phn_pyopenjtalk_jp/valid/text_shape.phn
|
83 |
+
- exp/svs_stats_raw_phn_pyopenjtalk_jp/valid/singing_shape
|
84 |
+
batch_type: sorted
|
85 |
+
valid_batch_type: null
|
86 |
+
fold_length:
|
87 |
+
- 150
|
88 |
+
- 240000
|
89 |
+
sort_in_batch: descending
|
90 |
+
shuffle_within_batch: false
|
91 |
+
sort_batch: descending
|
92 |
+
multiple_iterator: false
|
93 |
+
chunk_length: 500
|
94 |
+
chunk_shift_ratio: 0.5
|
95 |
+
num_cache_chunks: 1024
|
96 |
+
chunk_excluded_key_prefixes: []
|
97 |
+
chunk_default_fs: null
|
98 |
+
train_data_path_and_name_and_type:
|
99 |
+
- - dump/raw/tr_no_dev/text
|
100 |
+
- text
|
101 |
+
- text
|
102 |
+
- - dump/raw/tr_no_dev/wav.scp
|
103 |
+
- singing
|
104 |
+
- sound
|
105 |
+
- - dump/raw/tr_no_dev/label
|
106 |
+
- label
|
107 |
+
- duration
|
108 |
+
- - dump/raw/tr_no_dev/score.scp
|
109 |
+
- score
|
110 |
+
- score
|
111 |
+
valid_data_path_and_name_and_type:
|
112 |
+
- - dump/raw/dev/text
|
113 |
+
- text
|
114 |
+
- text
|
115 |
+
- - dump/raw/dev/wav.scp
|
116 |
+
- singing
|
117 |
+
- sound
|
118 |
+
- - dump/raw/dev/label
|
119 |
+
- label
|
120 |
+
- duration
|
121 |
+
- - dump/raw/dev/score.scp
|
122 |
+
- score
|
123 |
+
- score
|
124 |
+
allow_variable_data_keys: false
|
125 |
+
max_cache_size: 0.0
|
126 |
+
max_cache_fd: 32
|
127 |
+
allow_multi_rates: false
|
128 |
+
valid_max_cache_size: null
|
129 |
+
exclude_weight_decay: false
|
130 |
+
exclude_weight_decay_conf: {}
|
131 |
+
optim: adam
|
132 |
+
optim_conf:
|
133 |
+
lr: 0.001
|
134 |
+
eps: 1.0e-06
|
135 |
+
weight_decay: 0.0
|
136 |
+
scheduler: null
|
137 |
+
scheduler_conf: {}
|
138 |
+
token_list:
|
139 |
+
- <blank>
|
140 |
+
- <unk>
|
141 |
+
- pau
|
142 |
+
- a
|
143 |
+
- i
|
144 |
+
- o
|
145 |
+
- e
|
146 |
+
- u
|
147 |
+
- k
|
148 |
+
- n
|
149 |
+
- r
|
150 |
+
- t
|
151 |
+
- m
|
152 |
+
- d
|
153 |
+
- s
|
154 |
+
- N
|
155 |
+
- sh
|
156 |
+
- g
|
157 |
+
- y
|
158 |
+
- b
|
159 |
+
- w
|
160 |
+
- cl
|
161 |
+
- ts
|
162 |
+
- z
|
163 |
+
- ch
|
164 |
+
- j
|
165 |
+
- h
|
166 |
+
- f
|
167 |
+
- p
|
168 |
+
- ky
|
169 |
+
- ry
|
170 |
+
- hy
|
171 |
+
- py
|
172 |
+
- ny
|
173 |
+
- <sos/eos>
|
174 |
+
odim: null
|
175 |
+
model_conf: {}
|
176 |
+
use_preprocessor: true
|
177 |
+
token_type: phn
|
178 |
+
bpemodel: null
|
179 |
+
non_linguistic_symbols: null
|
180 |
+
cleaner: null
|
181 |
+
g2p: pyopenjtalk
|
182 |
+
fs: 24000
|
183 |
+
score_feats_extract: syllable_score_feats
|
184 |
+
score_feats_extract_conf:
|
185 |
+
fs: 24000
|
186 |
+
n_fft: 2048
|
187 |
+
win_length: 1200
|
188 |
+
hop_length: 300
|
189 |
+
feats_extract: fbank
|
190 |
+
feats_extract_conf:
|
191 |
+
n_fft: 2048
|
192 |
+
hop_length: 300
|
193 |
+
win_length: 1200
|
194 |
+
fs: 24000
|
195 |
+
fmin: 80
|
196 |
+
fmax: 7600
|
197 |
+
n_mels: 80
|
198 |
+
normalize: global_mvn
|
199 |
+
normalize_conf:
|
200 |
+
stats_file: exp/svs_stats_raw_phn_pyopenjtalk_jp/train/feats_stats.npz
|
201 |
+
svs: naive_rnn_dp
|
202 |
+
svs_conf:
|
203 |
+
midi_dim: 129
|
204 |
+
embed_dim: 512
|
205 |
+
duration_dim: 500
|
206 |
+
eprenet_conv_layers: 0
|
207 |
+
eprenet_conv_chans: 256
|
208 |
+
eprenet_conv_filts: 3
|
209 |
+
elayers: 3
|
210 |
+
eunits: 256
|
211 |
+
ebidirectional: true
|
212 |
+
midi_embed_integration_type: add
|
213 |
+
dlayers: 2
|
214 |
+
dunits: 256
|
215 |
+
dbidirectional: true
|
216 |
+
postnet_layers: 5
|
217 |
+
postnet_chans: 512
|
218 |
+
postnet_filts: 5
|
219 |
+
use_batch_norm: true
|
220 |
+
reduction_factor: 1
|
221 |
+
eprenet_dropout_rate: 0.2
|
222 |
+
edropout_rate: 0.1
|
223 |
+
ddropout_rate: 0.1
|
224 |
+
postnet_dropout_rate: 0.5
|
225 |
+
init_type: pytorch
|
226 |
+
use_masking: true
|
227 |
+
pitch_extract: dio
|
228 |
+
pitch_extract_conf:
|
229 |
+
use_token_averaged_f0: false
|
230 |
+
fs: 24000
|
231 |
+
n_fft: 2048
|
232 |
+
hop_length: 300
|
233 |
+
f0max: 800
|
234 |
+
f0min: 80
|
235 |
+
reduction_factor: 1
|
236 |
+
pitch_normalize: global_mvn
|
237 |
+
pitch_normalize_conf:
|
238 |
+
stats_file: exp/svs_stats_raw_phn_pyopenjtalk_jp/train/pitch_stats.npz
|
239 |
+
ying_extract: null
|
240 |
+
ying_extract_conf: {}
|
241 |
+
energy_extract: null
|
242 |
+
energy_extract_conf: {}
|
243 |
+
energy_normalize: null
|
244 |
+
energy_normalize_conf: {}
|
245 |
+
required:
|
246 |
+
- output_dir
|
247 |
+
- token_list
|
248 |
+
version: '202310'
|
249 |
+
distributed: false
|
exp/svs_train_naive_rnn_dp_raw_phn_pyopenjtalk_jp/images/backward_time.png
ADDED
exp/svs_train_naive_rnn_dp_raw_phn_pyopenjtalk_jp/images/clip.png
ADDED
exp/svs_train_naive_rnn_dp_raw_phn_pyopenjtalk_jp/images/duration_loss.png
ADDED
exp/svs_train_naive_rnn_dp_raw_phn_pyopenjtalk_jp/images/forward_time.png
ADDED
exp/svs_train_naive_rnn_dp_raw_phn_pyopenjtalk_jp/images/gpu_max_cached_mem_GB.png
ADDED
exp/svs_train_naive_rnn_dp_raw_phn_pyopenjtalk_jp/images/grad_norm.png
ADDED
exp/svs_train_naive_rnn_dp_raw_phn_pyopenjtalk_jp/images/iter_time.png
ADDED
exp/svs_train_naive_rnn_dp_raw_phn_pyopenjtalk_jp/images/l1_loss.png
ADDED
exp/svs_train_naive_rnn_dp_raw_phn_pyopenjtalk_jp/images/loss.png
ADDED
exp/svs_train_naive_rnn_dp_raw_phn_pyopenjtalk_jp/images/loss_scale.png
ADDED
exp/svs_train_naive_rnn_dp_raw_phn_pyopenjtalk_jp/images/optim0_lr0.png
ADDED
exp/svs_train_naive_rnn_dp_raw_phn_pyopenjtalk_jp/images/optim_step_time.png
ADDED
exp/svs_train_naive_rnn_dp_raw_phn_pyopenjtalk_jp/images/train_time.png
ADDED
meta.yaml
ADDED
@@ -0,0 +1,8 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
espnet: '202310'
|
2 |
+
files:
|
3 |
+
model_file: exp/svs_train_naive_rnn_dp_raw_phn_pyopenjtalk_jp/263epoch.pth
|
4 |
+
python: "3.9.16 (main, Mar 8 2023, 14:00:05) \n[GCC 11.2.0]"
|
5 |
+
timestamp: 1702989980.633902
|
6 |
+
torch: 1.13.1+cu117
|
7 |
+
yaml_files:
|
8 |
+
train_config: exp/svs_train_naive_rnn_dp_raw_phn_pyopenjtalk_jp/config.yaml
|