Commit
•
8372fea
1
Parent(s):
0c8d698
Update model
Browse files- README.md +327 -0
- exp/hubert_iter0_train_ssl_torchaudiohubert_base_960h_pretrain_it0_raw/config.yaml +257 -0
- exp/hubert_iter0_train_ssl_torchaudiohubert_base_960h_pretrain_it0_raw/images/acc_m.png +0 -0
- exp/hubert_iter0_train_ssl_torchaudiohubert_base_960h_pretrain_it0_raw/images/acc_u.png +0 -0
- exp/hubert_iter0_train_ssl_torchaudiohubert_base_960h_pretrain_it0_raw/images/backward_time.png +0 -0
- exp/hubert_iter0_train_ssl_torchaudiohubert_base_960h_pretrain_it0_raw/images/correct_m.png +0 -0
- exp/hubert_iter0_train_ssl_torchaudiohubert_base_960h_pretrain_it0_raw/images/correct_u.png +0 -0
- exp/hubert_iter0_train_ssl_torchaudiohubert_base_960h_pretrain_it0_raw/images/count_m.png +0 -0
- exp/hubert_iter0_train_ssl_torchaudiohubert_base_960h_pretrain_it0_raw/images/count_u.png +0 -0
- exp/hubert_iter0_train_ssl_torchaudiohubert_base_960h_pretrain_it0_raw/images/forward_time.png +0 -0
- exp/hubert_iter0_train_ssl_torchaudiohubert_base_960h_pretrain_it0_raw/images/gpu_max_cached_mem_GB.png +0 -0
- exp/hubert_iter0_train_ssl_torchaudiohubert_base_960h_pretrain_it0_raw/images/iter_time.png +0 -0
- exp/hubert_iter0_train_ssl_torchaudiohubert_base_960h_pretrain_it0_raw/images/loss.png +0 -0
- exp/hubert_iter0_train_ssl_torchaudiohubert_base_960h_pretrain_it0_raw/images/optim0_lr0.png +0 -0
- exp/hubert_iter0_train_ssl_torchaudiohubert_base_960h_pretrain_it0_raw/images/optim_step_time.png +0 -0
- exp/hubert_iter0_train_ssl_torchaudiohubert_base_960h_pretrain_it0_raw/images/train_time.png +0 -0
- exp/hubert_iter0_train_ssl_torchaudiohubert_base_960h_pretrain_it0_raw/valid.loss.ave.pth +3 -0
- exp/kmeans_iter0_mfcc_train_960_portion0.1/km_100.mdl +0 -0
- meta.yaml +8 -0
README.md
CHANGED
@@ -1,3 +1,330 @@
|
|
1 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2 |
license: cc-by-4.0
|
3 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
---
|
2 |
+
tags:
|
3 |
+
- espnet
|
4 |
+
- audio
|
5 |
+
- self-supervised-learning
|
6 |
+
language: en
|
7 |
+
datasets:
|
8 |
+
- librispeech
|
9 |
license: cc-by-4.0
|
10 |
---
|
11 |
+
|
12 |
+
## ESPnet2 SSL model
|
13 |
+
|
14 |
+
### `simpleoier/simpleoier_librispeech_hubert_iter0_train_ssl_torchaudiohubert_base_960h_pretrain_it0_raw`
|
15 |
+
|
16 |
+
This model was trained by simpleoier using librispeech 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 753f40d61813436d4e76660904d02eaed7a6649e
|
26 |
+
pip install -e .
|
27 |
+
cd egs2/librispeech/ssl1
|
28 |
+
./run.sh --skip_data_prep false --skip_train true --download_model simpleoier/simpleoier_librispeech_hubert_iter0_train_ssl_torchaudiohubert_base_960h_pretrain_it0_raw
|
29 |
+
```
|
30 |
+
|
31 |
+
|
32 |
+
|
33 |
+
## SSL config
|
34 |
+
|
35 |
+
<details><summary>expand</summary>
|
36 |
+
|
37 |
+
```
|
38 |
+
config: conf/tuning/train_ssl_torchaudiohubert_base_960h_pretrain_it0.yaml
|
39 |
+
print_config: false
|
40 |
+
log_level: INFO
|
41 |
+
dry_run: false
|
42 |
+
iterator_type: sequence
|
43 |
+
output_dir: exp/hubert_iter0_train_ssl_torchaudiohubert_base_960h_pretrain_it0_raw
|
44 |
+
ngpu: 1
|
45 |
+
seed: 0
|
46 |
+
num_workers: 64
|
47 |
+
num_att_plot: 3
|
48 |
+
dist_backend: nccl
|
49 |
+
dist_init_method: env://
|
50 |
+
dist_world_size: 8
|
51 |
+
dist_rank: 0
|
52 |
+
local_rank: 0
|
53 |
+
dist_master_addr: localhost
|
54 |
+
dist_master_port: 45091
|
55 |
+
dist_launcher: null
|
56 |
+
multiprocessing_distributed: true
|
57 |
+
unused_parameters: true
|
58 |
+
sharded_ddp: false
|
59 |
+
cudnn_enabled: true
|
60 |
+
cudnn_benchmark: false
|
61 |
+
cudnn_deterministic: true
|
62 |
+
collect_stats: false
|
63 |
+
write_collected_feats: false
|
64 |
+
max_epoch: 250
|
65 |
+
patience: null
|
66 |
+
val_scheduler_criterion:
|
67 |
+
- valid
|
68 |
+
- loss
|
69 |
+
early_stopping_criterion:
|
70 |
+
- valid
|
71 |
+
- loss
|
72 |
+
- min
|
73 |
+
best_model_criterion:
|
74 |
+
- - valid
|
75 |
+
- loss
|
76 |
+
- min
|
77 |
+
keep_nbest_models: 10
|
78 |
+
nbest_averaging_interval: 0
|
79 |
+
grad_clip: 5.0
|
80 |
+
grad_clip_type: 2.0
|
81 |
+
grad_noise: false
|
82 |
+
accum_grad: 2
|
83 |
+
no_forward_run: false
|
84 |
+
resume: true
|
85 |
+
train_dtype: float32
|
86 |
+
use_amp: true
|
87 |
+
log_interval: null
|
88 |
+
use_matplotlib: true
|
89 |
+
use_tensorboard: true
|
90 |
+
create_graph_in_tensorboard: false
|
91 |
+
use_wandb: false
|
92 |
+
wandb_project: null
|
93 |
+
wandb_id: null
|
94 |
+
wandb_entity: null
|
95 |
+
wandb_name: null
|
96 |
+
wandb_model_log_interval: -1
|
97 |
+
detect_anomaly: false
|
98 |
+
pretrain_path: null
|
99 |
+
init_param: []
|
100 |
+
ignore_init_mismatch: false
|
101 |
+
freeze_param: []
|
102 |
+
num_iters_per_epoch: null
|
103 |
+
batch_size: 20
|
104 |
+
valid_batch_size: null
|
105 |
+
batch_bins: 48000000
|
106 |
+
valid_batch_bins: null
|
107 |
+
train_shape_file:
|
108 |
+
- exp/hubert_iter0_stats_raw/train/speech_shape
|
109 |
+
- exp/hubert_iter0_stats_raw/train/text_shape.word
|
110 |
+
valid_shape_file:
|
111 |
+
- exp/hubert_iter0_stats_raw/valid/speech_shape
|
112 |
+
- exp/hubert_iter0_stats_raw/valid/text_shape.word
|
113 |
+
batch_type: numel
|
114 |
+
valid_batch_type: null
|
115 |
+
fold_length:
|
116 |
+
- 80000
|
117 |
+
- 400
|
118 |
+
sort_in_batch: descending
|
119 |
+
sort_batch: descending
|
120 |
+
multiple_iterator: false
|
121 |
+
chunk_length: 500
|
122 |
+
chunk_shift_ratio: 0.5
|
123 |
+
num_cache_chunks: 1024
|
124 |
+
train_data_path_and_name_and_type:
|
125 |
+
- - dump/raw/train_960/wav.scp
|
126 |
+
- speech
|
127 |
+
- sound
|
128 |
+
- - dump/raw/train_960/text.km.kmeans_iter0_mfcc_train_960_portion0.1
|
129 |
+
- text
|
130 |
+
- text
|
131 |
+
valid_data_path_and_name_and_type:
|
132 |
+
- - dump/raw/dev/wav.scp
|
133 |
+
- speech
|
134 |
+
- sound
|
135 |
+
- - dump/raw/dev/text.km.kmeans_iter0_mfcc_train_960_portion0.1
|
136 |
+
- text
|
137 |
+
- text
|
138 |
+
allow_variable_data_keys: false
|
139 |
+
max_cache_size: 0.0
|
140 |
+
max_cache_fd: 32
|
141 |
+
valid_max_cache_size: null
|
142 |
+
optim: adam
|
143 |
+
optim_conf:
|
144 |
+
lr: 0.0005
|
145 |
+
scheduler: warmuplr
|
146 |
+
scheduler_conf:
|
147 |
+
warmup_steps: 32000
|
148 |
+
token_list:
|
149 |
+
- '81'
|
150 |
+
- '5'
|
151 |
+
- '79'
|
152 |
+
- '84'
|
153 |
+
- '27'
|
154 |
+
- '35'
|
155 |
+
- '67'
|
156 |
+
- '56'
|
157 |
+
- '10'
|
158 |
+
- '99'
|
159 |
+
- '24'
|
160 |
+
- '3'
|
161 |
+
- '48'
|
162 |
+
- '8'
|
163 |
+
- '42'
|
164 |
+
- '16'
|
165 |
+
- '32'
|
166 |
+
- '31'
|
167 |
+
- '47'
|
168 |
+
- '43'
|
169 |
+
- '20'
|
170 |
+
- '73'
|
171 |
+
- '49'
|
172 |
+
- '86'
|
173 |
+
- '18'
|
174 |
+
- '64'
|
175 |
+
- '34'
|
176 |
+
- '59'
|
177 |
+
- '95'
|
178 |
+
- '0'
|
179 |
+
- '52'
|
180 |
+
- '44'
|
181 |
+
- '61'
|
182 |
+
- '57'
|
183 |
+
- '30'
|
184 |
+
- '1'
|
185 |
+
- '93'
|
186 |
+
- '6'
|
187 |
+
- '69'
|
188 |
+
- '19'
|
189 |
+
- '7'
|
190 |
+
- '65'
|
191 |
+
- '28'
|
192 |
+
- '89'
|
193 |
+
- '2'
|
194 |
+
- '96'
|
195 |
+
- '91'
|
196 |
+
- '72'
|
197 |
+
- '38'
|
198 |
+
- '78'
|
199 |
+
- '26'
|
200 |
+
- '13'
|
201 |
+
- '39'
|
202 |
+
- '94'
|
203 |
+
- '4'
|
204 |
+
- '88'
|
205 |
+
- '85'
|
206 |
+
- '51'
|
207 |
+
- '82'
|
208 |
+
- '41'
|
209 |
+
- '50'
|
210 |
+
- '21'
|
211 |
+
- '80'
|
212 |
+
- '97'
|
213 |
+
- '87'
|
214 |
+
- '25'
|
215 |
+
- '54'
|
216 |
+
- '12'
|
217 |
+
- '40'
|
218 |
+
- '60'
|
219 |
+
- '29'
|
220 |
+
- '11'
|
221 |
+
- '53'
|
222 |
+
- '71'
|
223 |
+
- '83'
|
224 |
+
- '74'
|
225 |
+
- '68'
|
226 |
+
- '55'
|
227 |
+
- '62'
|
228 |
+
- '76'
|
229 |
+
- '45'
|
230 |
+
- '75'
|
231 |
+
- '92'
|
232 |
+
- '46'
|
233 |
+
- '36'
|
234 |
+
- '66'
|
235 |
+
- '22'
|
236 |
+
- '77'
|
237 |
+
- '23'
|
238 |
+
- '63'
|
239 |
+
- '37'
|
240 |
+
- '58'
|
241 |
+
- '33'
|
242 |
+
- '15'
|
243 |
+
- '17'
|
244 |
+
- '90'
|
245 |
+
- '98'
|
246 |
+
- '14'
|
247 |
+
- '70'
|
248 |
+
- '9'
|
249 |
+
- <unk>
|
250 |
+
- <sos/eos>
|
251 |
+
init: null
|
252 |
+
collate_fn_conf:
|
253 |
+
label_downsampling: 2
|
254 |
+
pad: false
|
255 |
+
rand_crop: true
|
256 |
+
input_size: 1
|
257 |
+
num_classes: 100
|
258 |
+
use_preprocessor: true
|
259 |
+
token_type: word
|
260 |
+
bpemodel: null
|
261 |
+
non_linguistic_symbols: null
|
262 |
+
cleaner: null
|
263 |
+
g2p: null
|
264 |
+
speech_volume_normalize: null
|
265 |
+
rir_scp: null
|
266 |
+
rir_apply_prob: 1.0
|
267 |
+
noise_scp: null
|
268 |
+
noise_apply_prob: 1.0
|
269 |
+
noise_db_range: '13_15'
|
270 |
+
pred_masked_weight: 1.0
|
271 |
+
pred_nomask_weight: 0.0
|
272 |
+
loss_weights: 0.0
|
273 |
+
frontend: null
|
274 |
+
frontend_conf: {}
|
275 |
+
specaug: null
|
276 |
+
specaug_conf: {}
|
277 |
+
normalize: null
|
278 |
+
normalize_conf: {}
|
279 |
+
preencoder: null
|
280 |
+
preencoder_conf: {}
|
281 |
+
encoder: torchaudio_hubert
|
282 |
+
encoder_conf:
|
283 |
+
encoder_projection_dropout: 0.1
|
284 |
+
encoder_attention_dropout: 0.1
|
285 |
+
encoder_ff_interm_dropout: 0.0
|
286 |
+
encoder_dropout: 0.1
|
287 |
+
encoder_layer_drop: 0.05
|
288 |
+
model: torchaudio
|
289 |
+
model_conf: {}
|
290 |
+
required:
|
291 |
+
- output_dir
|
292 |
+
- token_list
|
293 |
+
version: '202209'
|
294 |
+
distributed: true
|
295 |
+
```
|
296 |
+
|
297 |
+
</details>
|
298 |
+
|
299 |
+
|
300 |
+
|
301 |
+
### Citing ESPnet
|
302 |
+
|
303 |
+
```BibTex
|
304 |
+
@inproceedings{watanabe2018espnet,
|
305 |
+
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},
|
306 |
+
title={{ESPnet}: End-to-End Speech Processing Toolkit},
|
307 |
+
year={2018},
|
308 |
+
booktitle={Proceedings of Interspeech},
|
309 |
+
pages={2207--2211},
|
310 |
+
doi={10.21437/Interspeech.2018-1456},
|
311 |
+
url={http://dx.doi.org/10.21437/Interspeech.2018-1456}
|
312 |
+
}
|
313 |
+
|
314 |
+
|
315 |
+
|
316 |
+
|
317 |
+
```
|
318 |
+
|
319 |
+
or arXiv:
|
320 |
+
|
321 |
+
```bibtex
|
322 |
+
@misc{watanabe2018espnet,
|
323 |
+
title={ESPnet: End-to-End Speech Processing Toolkit},
|
324 |
+
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},
|
325 |
+
year={2018},
|
326 |
+
eprint={1804.00015},
|
327 |
+
archivePrefix={arXiv},
|
328 |
+
primaryClass={cs.CL}
|
329 |
+
}
|
330 |
+
```
|
exp/hubert_iter0_train_ssl_torchaudiohubert_base_960h_pretrain_it0_raw/config.yaml
ADDED
@@ -0,0 +1,257 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
config: conf/tuning/train_ssl_torchaudiohubert_base_960h_pretrain_it0.yaml
|
2 |
+
print_config: false
|
3 |
+
log_level: INFO
|
4 |
+
dry_run: false
|
5 |
+
iterator_type: sequence
|
6 |
+
output_dir: exp/hubert_iter0_train_ssl_torchaudiohubert_base_960h_pretrain_it0_raw
|
7 |
+
ngpu: 1
|
8 |
+
seed: 0
|
9 |
+
num_workers: 64
|
10 |
+
num_att_plot: 3
|
11 |
+
dist_backend: nccl
|
12 |
+
dist_init_method: env://
|
13 |
+
dist_world_size: 8
|
14 |
+
dist_rank: 0
|
15 |
+
local_rank: 0
|
16 |
+
dist_master_addr: localhost
|
17 |
+
dist_master_port: 45091
|
18 |
+
dist_launcher: null
|
19 |
+
multiprocessing_distributed: true
|
20 |
+
unused_parameters: true
|
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: 250
|
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 |
+
keep_nbest_models: 10
|
41 |
+
nbest_averaging_interval: 0
|
42 |
+
grad_clip: 5.0
|
43 |
+
grad_clip_type: 2.0
|
44 |
+
grad_noise: false
|
45 |
+
accum_grad: 2
|
46 |
+
no_forward_run: false
|
47 |
+
resume: true
|
48 |
+
train_dtype: float32
|
49 |
+
use_amp: true
|
50 |
+
log_interval: null
|
51 |
+
use_matplotlib: true
|
52 |
+
use_tensorboard: true
|
53 |
+
create_graph_in_tensorboard: false
|
54 |
+
use_wandb: false
|
55 |
+
wandb_project: null
|
56 |
+
wandb_id: null
|
57 |
+
wandb_entity: null
|
58 |
+
wandb_name: null
|
59 |
+
wandb_model_log_interval: -1
|
60 |
+
detect_anomaly: false
|
61 |
+
pretrain_path: null
|
62 |
+
init_param: []
|
63 |
+
ignore_init_mismatch: false
|
64 |
+
freeze_param: []
|
65 |
+
num_iters_per_epoch: null
|
66 |
+
batch_size: 20
|
67 |
+
valid_batch_size: null
|
68 |
+
batch_bins: 48000000
|
69 |
+
valid_batch_bins: null
|
70 |
+
train_shape_file:
|
71 |
+
- exp/hubert_iter0_stats_raw/train/speech_shape
|
72 |
+
- exp/hubert_iter0_stats_raw/train/text_shape.word
|
73 |
+
valid_shape_file:
|
74 |
+
- exp/hubert_iter0_stats_raw/valid/speech_shape
|
75 |
+
- exp/hubert_iter0_stats_raw/valid/text_shape.word
|
76 |
+
batch_type: numel
|
77 |
+
valid_batch_type: null
|
78 |
+
fold_length:
|
79 |
+
- 80000
|
80 |
+
- 400
|
81 |
+
sort_in_batch: descending
|
82 |
+
sort_batch: descending
|
83 |
+
multiple_iterator: false
|
84 |
+
chunk_length: 500
|
85 |
+
chunk_shift_ratio: 0.5
|
86 |
+
num_cache_chunks: 1024
|
87 |
+
train_data_path_and_name_and_type:
|
88 |
+
- - dump/raw/train_960/wav.scp
|
89 |
+
- speech
|
90 |
+
- sound
|
91 |
+
- - dump/raw/train_960/text.km.kmeans_iter0_mfcc_train_960_portion0.1
|
92 |
+
- text
|
93 |
+
- text
|
94 |
+
valid_data_path_and_name_and_type:
|
95 |
+
- - dump/raw/dev/wav.scp
|
96 |
+
- speech
|
97 |
+
- sound
|
98 |
+
- - dump/raw/dev/text.km.kmeans_iter0_mfcc_train_960_portion0.1
|
99 |
+
- text
|
100 |
+
- text
|
101 |
+
allow_variable_data_keys: false
|
102 |
+
max_cache_size: 0.0
|
103 |
+
max_cache_fd: 32
|
104 |
+
valid_max_cache_size: null
|
105 |
+
optim: adam
|
106 |
+
optim_conf:
|
107 |
+
lr: 0.0005
|
108 |
+
scheduler: warmuplr
|
109 |
+
scheduler_conf:
|
110 |
+
warmup_steps: 32000
|
111 |
+
token_list:
|
112 |
+
- '81'
|
113 |
+
- '5'
|
114 |
+
- '79'
|
115 |
+
- '84'
|
116 |
+
- '27'
|
117 |
+
- '35'
|
118 |
+
- '67'
|
119 |
+
- '56'
|
120 |
+
- '10'
|
121 |
+
- '99'
|
122 |
+
- '24'
|
123 |
+
- '3'
|
124 |
+
- '48'
|
125 |
+
- '8'
|
126 |
+
- '42'
|
127 |
+
- '16'
|
128 |
+
- '32'
|
129 |
+
- '31'
|
130 |
+
- '47'
|
131 |
+
- '43'
|
132 |
+
- '20'
|
133 |
+
- '73'
|
134 |
+
- '49'
|
135 |
+
- '86'
|
136 |
+
- '18'
|
137 |
+
- '64'
|
138 |
+
- '34'
|
139 |
+
- '59'
|
140 |
+
- '95'
|
141 |
+
- '0'
|
142 |
+
- '52'
|
143 |
+
- '44'
|
144 |
+
- '61'
|
145 |
+
- '57'
|
146 |
+
- '30'
|
147 |
+
- '1'
|
148 |
+
- '93'
|
149 |
+
- '6'
|
150 |
+
- '69'
|
151 |
+
- '19'
|
152 |
+
- '7'
|
153 |
+
- '65'
|
154 |
+
- '28'
|
155 |
+
- '89'
|
156 |
+
- '2'
|
157 |
+
- '96'
|
158 |
+
- '91'
|
159 |
+
- '72'
|
160 |
+
- '38'
|
161 |
+
- '78'
|
162 |
+
- '26'
|
163 |
+
- '13'
|
164 |
+
- '39'
|
165 |
+
- '94'
|
166 |
+
- '4'
|
167 |
+
- '88'
|
168 |
+
- '85'
|
169 |
+
- '51'
|
170 |
+
- '82'
|
171 |
+
- '41'
|
172 |
+
- '50'
|
173 |
+
- '21'
|
174 |
+
- '80'
|
175 |
+
- '97'
|
176 |
+
- '87'
|
177 |
+
- '25'
|
178 |
+
- '54'
|
179 |
+
- '12'
|
180 |
+
- '40'
|
181 |
+
- '60'
|
182 |
+
- '29'
|
183 |
+
- '11'
|
184 |
+
- '53'
|
185 |
+
- '71'
|
186 |
+
- '83'
|
187 |
+
- '74'
|
188 |
+
- '68'
|
189 |
+
- '55'
|
190 |
+
- '62'
|
191 |
+
- '76'
|
192 |
+
- '45'
|
193 |
+
- '75'
|
194 |
+
- '92'
|
195 |
+
- '46'
|
196 |
+
- '36'
|
197 |
+
- '66'
|
198 |
+
- '22'
|
199 |
+
- '77'
|
200 |
+
- '23'
|
201 |
+
- '63'
|
202 |
+
- '37'
|
203 |
+
- '58'
|
204 |
+
- '33'
|
205 |
+
- '15'
|
206 |
+
- '17'
|
207 |
+
- '90'
|
208 |
+
- '98'
|
209 |
+
- '14'
|
210 |
+
- '70'
|
211 |
+
- '9'
|
212 |
+
- <unk>
|
213 |
+
- <sos/eos>
|
214 |
+
init: null
|
215 |
+
collate_fn_conf:
|
216 |
+
label_downsampling: 2
|
217 |
+
pad: false
|
218 |
+
rand_crop: true
|
219 |
+
input_size: 1
|
220 |
+
num_classes: 100
|
221 |
+
use_preprocessor: true
|
222 |
+
token_type: word
|
223 |
+
bpemodel: null
|
224 |
+
non_linguistic_symbols: null
|
225 |
+
cleaner: null
|
226 |
+
g2p: null
|
227 |
+
speech_volume_normalize: null
|
228 |
+
rir_scp: null
|
229 |
+
rir_apply_prob: 1.0
|
230 |
+
noise_scp: null
|
231 |
+
noise_apply_prob: 1.0
|
232 |
+
noise_db_range: '13_15'
|
233 |
+
pred_masked_weight: 1.0
|
234 |
+
pred_nomask_weight: 0.0
|
235 |
+
loss_weights: 0.0
|
236 |
+
frontend: null
|
237 |
+
frontend_conf: {}
|
238 |
+
specaug: null
|
239 |
+
specaug_conf: {}
|
240 |
+
normalize: null
|
241 |
+
normalize_conf: {}
|
242 |
+
preencoder: null
|
243 |
+
preencoder_conf: {}
|
244 |
+
encoder: torchaudio_hubert
|
245 |
+
encoder_conf:
|
246 |
+
encoder_projection_dropout: 0.1
|
247 |
+
encoder_attention_dropout: 0.1
|
248 |
+
encoder_ff_interm_dropout: 0.0
|
249 |
+
encoder_dropout: 0.1
|
250 |
+
encoder_layer_drop: 0.05
|
251 |
+
model: torchaudio
|
252 |
+
model_conf: {}
|
253 |
+
required:
|
254 |
+
- output_dir
|
255 |
+
- token_list
|
256 |
+
version: '202209'
|
257 |
+
distributed: true
|
exp/hubert_iter0_train_ssl_torchaudiohubert_base_960h_pretrain_it0_raw/images/acc_m.png
ADDED
![]() |
exp/hubert_iter0_train_ssl_torchaudiohubert_base_960h_pretrain_it0_raw/images/acc_u.png
ADDED
![]() |
exp/hubert_iter0_train_ssl_torchaudiohubert_base_960h_pretrain_it0_raw/images/backward_time.png
ADDED
![]() |
exp/hubert_iter0_train_ssl_torchaudiohubert_base_960h_pretrain_it0_raw/images/correct_m.png
ADDED
![]() |
exp/hubert_iter0_train_ssl_torchaudiohubert_base_960h_pretrain_it0_raw/images/correct_u.png
ADDED
![]() |
exp/hubert_iter0_train_ssl_torchaudiohubert_base_960h_pretrain_it0_raw/images/count_m.png
ADDED
![]() |
exp/hubert_iter0_train_ssl_torchaudiohubert_base_960h_pretrain_it0_raw/images/count_u.png
ADDED
![]() |
exp/hubert_iter0_train_ssl_torchaudiohubert_base_960h_pretrain_it0_raw/images/forward_time.png
ADDED
![]() |
exp/hubert_iter0_train_ssl_torchaudiohubert_base_960h_pretrain_it0_raw/images/gpu_max_cached_mem_GB.png
ADDED
![]() |
exp/hubert_iter0_train_ssl_torchaudiohubert_base_960h_pretrain_it0_raw/images/iter_time.png
ADDED
![]() |
exp/hubert_iter0_train_ssl_torchaudiohubert_base_960h_pretrain_it0_raw/images/loss.png
ADDED
![]() |
exp/hubert_iter0_train_ssl_torchaudiohubert_base_960h_pretrain_it0_raw/images/optim0_lr0.png
ADDED
![]() |
exp/hubert_iter0_train_ssl_torchaudiohubert_base_960h_pretrain_it0_raw/images/optim_step_time.png
ADDED
![]() |
exp/hubert_iter0_train_ssl_torchaudiohubert_base_960h_pretrain_it0_raw/images/train_time.png
ADDED
![]() |
exp/hubert_iter0_train_ssl_torchaudiohubert_base_960h_pretrain_it0_raw/valid.loss.ave.pth
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:313bb75a1a1c35c95eb6ee62f0773be1c6ba58798828a335f00b82e3af14022c
|
3 |
+
size 378481137
|
exp/kmeans_iter0_mfcc_train_960_portion0.1/km_100.mdl
ADDED
Binary file (16.9 kB). View file
|
|
meta.yaml
ADDED
@@ -0,0 +1,8 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
espnet: '202209'
|
2 |
+
files:
|
3 |
+
model_file: exp/hubert_iter0_train_ssl_torchaudiohubert_base_960h_pretrain_it0_raw/valid.loss.ave.pth
|
4 |
+
python: "3.9.15 (main, Nov 24 2022, 14:31:59) \n[GCC 11.2.0]"
|
5 |
+
timestamp: 1672839599.248442
|
6 |
+
torch: 1.13.0
|
7 |
+
yaml_files:
|
8 |
+
train_config: exp/hubert_iter0_train_ssl_torchaudiohubert_base_960h_pretrain_it0_raw/config.yaml
|