add model files
Browse files- README.md +371 -0
- data/nlsyms.txt +3 -0
- exp/asr_stats_raw_en_char_sp/train/feats_stats.npz +3 -0
- exp/asr_train_asr_conformer_e12_linear2048_raw_en_char_sp/RESULTS.md +57 -0
- exp/asr_train_asr_conformer_e12_linear2048_raw_en_char_sp/config.yaml +266 -0
- exp/asr_train_asr_conformer_e12_linear2048_raw_en_char_sp/images/acc.png +0 -0
- exp/asr_train_asr_conformer_e12_linear2048_raw_en_char_sp/images/backward_time.png +0 -0
- exp/asr_train_asr_conformer_e12_linear2048_raw_en_char_sp/images/cer.png +0 -0
- exp/asr_train_asr_conformer_e12_linear2048_raw_en_char_sp/images/cer_ctc.png +0 -0
- exp/asr_train_asr_conformer_e12_linear2048_raw_en_char_sp/images/forward_time.png +0 -0
- exp/asr_train_asr_conformer_e12_linear2048_raw_en_char_sp/images/gpu_max_cached_mem_GB.png +0 -0
- exp/asr_train_asr_conformer_e12_linear2048_raw_en_char_sp/images/iter_time.png +0 -0
- exp/asr_train_asr_conformer_e12_linear2048_raw_en_char_sp/images/loss.png +0 -0
- exp/asr_train_asr_conformer_e12_linear2048_raw_en_char_sp/images/loss_att.png +0 -0
- exp/asr_train_asr_conformer_e12_linear2048_raw_en_char_sp/images/loss_ctc.png +0 -0
- exp/asr_train_asr_conformer_e12_linear2048_raw_en_char_sp/images/optim0_lr0.png +0 -0
- exp/asr_train_asr_conformer_e12_linear2048_raw_en_char_sp/images/optim_step_time.png +0 -0
- exp/asr_train_asr_conformer_e12_linear2048_raw_en_char_sp/images/train_time.png +0 -0
- exp/asr_train_asr_conformer_e12_linear2048_raw_en_char_sp/images/wer.png +0 -0
- exp/asr_train_asr_conformer_e12_linear2048_raw_en_char_sp/valid.acc.ave_10best.pth +3 -0
- exp/lm_train_lm_transformer_en_char/12epoch.pth +3 -0
- exp/lm_train_lm_transformer_en_char/config.yaml +189 -0
- exp/lm_train_lm_transformer_en_char/images/backward_time.png +0 -0
- exp/lm_train_lm_transformer_en_char/images/forward_time.png +0 -0
- exp/lm_train_lm_transformer_en_char/images/gpu_max_cached_mem_GB.png +0 -0
- exp/lm_train_lm_transformer_en_char/images/iter_time.png +0 -0
- exp/lm_train_lm_transformer_en_char/images/loss.png +0 -0
- exp/lm_train_lm_transformer_en_char/images/optim0_lr0.png +0 -0
- exp/lm_train_lm_transformer_en_char/images/optim_step_time.png +0 -0
- exp/lm_train_lm_transformer_en_char/images/train_time.png +0 -0
- exp/lm_train_lm_transformer_en_char/perplexity_test/ppl +1 -0
- meta.yaml +10 -0
README.md
ADDED
|
@@ -0,0 +1,371 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
tags:
|
| 3 |
+
- espnet
|
| 4 |
+
- audio
|
| 5 |
+
- automatic-speech-recognition
|
| 6 |
+
language: en
|
| 7 |
+
datasets:
|
| 8 |
+
- chime4
|
| 9 |
+
license: cc-by-4.0
|
| 10 |
+
---
|
| 11 |
+
|
| 12 |
+
## ESPnet2 ASR model
|
| 13 |
+
|
| 14 |
+
### `pyf98/chime4_conformer_e12_linear2048`
|
| 15 |
+
|
| 16 |
+
This model was trained by Yifan Peng using chime4 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 ad91279f0108d54bd22abe29671b376f048822c5
|
| 26 |
+
pip install -e .
|
| 27 |
+
cd egs2/chime4/asr1
|
| 28 |
+
./run.sh --skip_data_prep false --skip_train true --download_model pyf98/chime4_conformer_e12_linear2048
|
| 29 |
+
```
|
| 30 |
+
|
| 31 |
+
<!-- Generated by scripts/utils/show_asr_result.sh -->
|
| 32 |
+
# RESULTS
|
| 33 |
+
## Environments
|
| 34 |
+
- date: `Wed Dec 28 20:41:40 EST 2022`
|
| 35 |
+
- python version: `3.9.15 (main, Nov 24 2022, 14:31:59) [GCC 11.2.0]`
|
| 36 |
+
- espnet version: `espnet 202211`
|
| 37 |
+
- pytorch version: `pytorch 1.12.1`
|
| 38 |
+
- Git hash: `ad91279f0108d54bd22abe29671b376f048822c5`
|
| 39 |
+
- Commit date: `Wed Dec 28 20:15:42 2022 -0500`
|
| 40 |
+
|
| 41 |
+
## asr_train_asr_conformer_e12_linear2048_raw_en_char_sp
|
| 42 |
+
### WER
|
| 43 |
+
|
| 44 |
+
|dataset|Snt|Wrd|Corr|Sub|Del|Ins|Err|S.Err|
|
| 45 |
+
|---|---|---|---|---|---|---|---|---|
|
| 46 |
+
|decode_asr_lm_lm_train_lm_transformer_en_char_valid.loss.ave_asr_model_valid.acc.ave/dt05_real_beamformit_5mics|1640|27119|93.3|5.4|1.3|0.5|7.3|55.6|
|
| 47 |
+
|decode_asr_lm_lm_train_lm_transformer_en_char_valid.loss.ave_asr_model_valid.acc.ave/dt05_simu_beamformit_5mics|1640|27120|91.7|6.7|1.6|0.9|9.1|62.0|
|
| 48 |
+
|decode_asr_lm_lm_train_lm_transformer_en_char_valid.loss.ave_asr_model_valid.acc.ave/et05_real_beamformit_5mics|1320|21409|89.2|8.9|1.9|1.1|12.0|64.5|
|
| 49 |
+
|decode_asr_lm_lm_train_lm_transformer_en_char_valid.loss.ave_asr_model_valid.acc.ave/et05_simu_beamformit_5mics|1320|21416|87.8|9.6|2.6|1.4|13.6|68.1|
|
| 50 |
+
|
| 51 |
+
### CER
|
| 52 |
+
|
| 53 |
+
|dataset|Snt|Wrd|Corr|Sub|Del|Ins|Err|S.Err|
|
| 54 |
+
|---|---|---|---|---|---|---|---|---|
|
| 55 |
+
|decode_asr_lm_lm_train_lm_transformer_en_char_valid.loss.ave_asr_model_valid.acc.ave/dt05_real_beamformit_5mics|1640|160390|97.2|1.5|1.3|0.7|3.5|55.6|
|
| 56 |
+
|decode_asr_lm_lm_train_lm_transformer_en_char_valid.loss.ave_asr_model_valid.acc.ave/dt05_simu_beamformit_5mics|1640|160400|96.3|2.0|1.7|1.0|4.7|62.0|
|
| 57 |
+
|decode_asr_lm_lm_train_lm_transformer_en_char_valid.loss.ave_asr_model_valid.acc.ave/et05_real_beamformit_5mics|1320|126796|95.1|2.8|2.1|1.2|6.1|64.6|
|
| 58 |
+
|decode_asr_lm_lm_train_lm_transformer_en_char_valid.loss.ave_asr_model_valid.acc.ave/et05_simu_beamformit_5mics|1320|126812|94.0|3.1|3.0|1.6|7.7|68.1|
|
| 59 |
+
|
| 60 |
+
### TER
|
| 61 |
+
|
| 62 |
+
|dataset|Snt|Wrd|Corr|Sub|Del|Ins|Err|S.Err|
|
| 63 |
+
|---|---|---|---|---|---|---|---|---|
|
| 64 |
+
|
| 65 |
+
## ASR config
|
| 66 |
+
|
| 67 |
+
<details><summary>expand</summary>
|
| 68 |
+
|
| 69 |
+
```
|
| 70 |
+
config: conf/tuning/train_asr_conformer_e12_linear2048.yaml
|
| 71 |
+
print_config: false
|
| 72 |
+
log_level: INFO
|
| 73 |
+
dry_run: false
|
| 74 |
+
iterator_type: sequence
|
| 75 |
+
output_dir: exp/asr_train_asr_conformer_e12_linear2048_raw_en_char_sp
|
| 76 |
+
ngpu: 1
|
| 77 |
+
seed: 2022
|
| 78 |
+
num_workers: 4
|
| 79 |
+
num_att_plot: 3
|
| 80 |
+
dist_backend: nccl
|
| 81 |
+
dist_init_method: env://
|
| 82 |
+
dist_world_size: 2
|
| 83 |
+
dist_rank: 0
|
| 84 |
+
local_rank: 0
|
| 85 |
+
dist_master_addr: localhost
|
| 86 |
+
dist_master_port: 45069
|
| 87 |
+
dist_launcher: null
|
| 88 |
+
multiprocessing_distributed: true
|
| 89 |
+
unused_parameters: false
|
| 90 |
+
sharded_ddp: false
|
| 91 |
+
cudnn_enabled: true
|
| 92 |
+
cudnn_benchmark: false
|
| 93 |
+
cudnn_deterministic: true
|
| 94 |
+
collect_stats: false
|
| 95 |
+
write_collected_feats: false
|
| 96 |
+
max_epoch: 50
|
| 97 |
+
patience: null
|
| 98 |
+
val_scheduler_criterion:
|
| 99 |
+
- valid
|
| 100 |
+
- loss
|
| 101 |
+
early_stopping_criterion:
|
| 102 |
+
- valid
|
| 103 |
+
- loss
|
| 104 |
+
- min
|
| 105 |
+
best_model_criterion:
|
| 106 |
+
- - valid
|
| 107 |
+
- acc
|
| 108 |
+
- max
|
| 109 |
+
keep_nbest_models: 10
|
| 110 |
+
nbest_averaging_interval: 0
|
| 111 |
+
grad_clip: 5.0
|
| 112 |
+
grad_clip_type: 2.0
|
| 113 |
+
grad_noise: false
|
| 114 |
+
accum_grad: 1
|
| 115 |
+
no_forward_run: false
|
| 116 |
+
resume: true
|
| 117 |
+
train_dtype: float32
|
| 118 |
+
use_amp: true
|
| 119 |
+
log_interval: null
|
| 120 |
+
use_matplotlib: true
|
| 121 |
+
use_tensorboard: true
|
| 122 |
+
create_graph_in_tensorboard: false
|
| 123 |
+
use_wandb: false
|
| 124 |
+
wandb_project: null
|
| 125 |
+
wandb_id: null
|
| 126 |
+
wandb_entity: null
|
| 127 |
+
wandb_name: null
|
| 128 |
+
wandb_model_log_interval: -1
|
| 129 |
+
detect_anomaly: false
|
| 130 |
+
pretrain_path: null
|
| 131 |
+
init_param: []
|
| 132 |
+
ignore_init_mismatch: false
|
| 133 |
+
freeze_param: []
|
| 134 |
+
num_iters_per_epoch: null
|
| 135 |
+
batch_size: 20
|
| 136 |
+
valid_batch_size: null
|
| 137 |
+
batch_bins: 15000000
|
| 138 |
+
valid_batch_bins: null
|
| 139 |
+
train_shape_file:
|
| 140 |
+
- exp/asr_stats_raw_en_char_sp/train/speech_shape
|
| 141 |
+
- exp/asr_stats_raw_en_char_sp/train/text_shape.char
|
| 142 |
+
valid_shape_file:
|
| 143 |
+
- exp/asr_stats_raw_en_char_sp/valid/speech_shape
|
| 144 |
+
- exp/asr_stats_raw_en_char_sp/valid/text_shape.char
|
| 145 |
+
batch_type: numel
|
| 146 |
+
valid_batch_type: null
|
| 147 |
+
fold_length:
|
| 148 |
+
- 80000
|
| 149 |
+
- 150
|
| 150 |
+
sort_in_batch: descending
|
| 151 |
+
sort_batch: descending
|
| 152 |
+
multiple_iterator: false
|
| 153 |
+
chunk_length: 500
|
| 154 |
+
chunk_shift_ratio: 0.5
|
| 155 |
+
num_cache_chunks: 1024
|
| 156 |
+
train_data_path_and_name_and_type:
|
| 157 |
+
- - dump/raw/tr05_multi_noisy_si284_sp/wav.scp
|
| 158 |
+
- speech
|
| 159 |
+
- kaldi_ark
|
| 160 |
+
- - dump/raw/tr05_multi_noisy_si284_sp/text
|
| 161 |
+
- text
|
| 162 |
+
- text
|
| 163 |
+
valid_data_path_and_name_and_type:
|
| 164 |
+
- - dump/raw/dt05_multi_isolated_1ch_track/wav.scp
|
| 165 |
+
- speech
|
| 166 |
+
- kaldi_ark
|
| 167 |
+
- - dump/raw/dt05_multi_isolated_1ch_track/text
|
| 168 |
+
- text
|
| 169 |
+
- text
|
| 170 |
+
allow_variable_data_keys: false
|
| 171 |
+
max_cache_size: 0.0
|
| 172 |
+
max_cache_fd: 32
|
| 173 |
+
valid_max_cache_size: null
|
| 174 |
+
optim: adam
|
| 175 |
+
optim_conf:
|
| 176 |
+
lr: 0.001
|
| 177 |
+
weight_decay: 1.0e-06
|
| 178 |
+
scheduler: warmuplr
|
| 179 |
+
scheduler_conf:
|
| 180 |
+
warmup_steps: 25000
|
| 181 |
+
token_list:
|
| 182 |
+
- <blank>
|
| 183 |
+
- <unk>
|
| 184 |
+
- <space>
|
| 185 |
+
- E
|
| 186 |
+
- T
|
| 187 |
+
- A
|
| 188 |
+
- N
|
| 189 |
+
- I
|
| 190 |
+
- O
|
| 191 |
+
- S
|
| 192 |
+
- R
|
| 193 |
+
- H
|
| 194 |
+
- L
|
| 195 |
+
- D
|
| 196 |
+
- C
|
| 197 |
+
- U
|
| 198 |
+
- M
|
| 199 |
+
- P
|
| 200 |
+
- F
|
| 201 |
+
- G
|
| 202 |
+
- Y
|
| 203 |
+
- W
|
| 204 |
+
- B
|
| 205 |
+
- V
|
| 206 |
+
- K
|
| 207 |
+
- .
|
| 208 |
+
- X
|
| 209 |
+
- ''''
|
| 210 |
+
- J
|
| 211 |
+
- Q
|
| 212 |
+
- Z
|
| 213 |
+
- ','
|
| 214 |
+
- '-'
|
| 215 |
+
- '"'
|
| 216 |
+
- <NOISE>
|
| 217 |
+
- '*'
|
| 218 |
+
- ':'
|
| 219 |
+
- (
|
| 220 |
+
- )
|
| 221 |
+
- '?'
|
| 222 |
+
- '&'
|
| 223 |
+
- ;
|
| 224 |
+
- '!'
|
| 225 |
+
- /
|
| 226 |
+
- '{'
|
| 227 |
+
- '}'
|
| 228 |
+
- '1'
|
| 229 |
+
- '2'
|
| 230 |
+
- '0'
|
| 231 |
+
- $
|
| 232 |
+
- '8'
|
| 233 |
+
- '9'
|
| 234 |
+
- '6'
|
| 235 |
+
- '3'
|
| 236 |
+
- '5'
|
| 237 |
+
- '7'
|
| 238 |
+
- '4'
|
| 239 |
+
- '~'
|
| 240 |
+
- '`'
|
| 241 |
+
- _
|
| 242 |
+
- <*IN*>
|
| 243 |
+
- <*MR.*>
|
| 244 |
+
- \
|
| 245 |
+
- ^
|
| 246 |
+
- <sos/eos>
|
| 247 |
+
init: null
|
| 248 |
+
input_size: null
|
| 249 |
+
ctc_conf:
|
| 250 |
+
dropout_rate: 0.0
|
| 251 |
+
ctc_type: builtin
|
| 252 |
+
reduce: true
|
| 253 |
+
ignore_nan_grad: null
|
| 254 |
+
zero_infinity: true
|
| 255 |
+
joint_net_conf: null
|
| 256 |
+
use_preprocessor: true
|
| 257 |
+
token_type: char
|
| 258 |
+
bpemodel: null
|
| 259 |
+
non_linguistic_symbols: data/nlsyms.txt
|
| 260 |
+
cleaner: null
|
| 261 |
+
g2p: null
|
| 262 |
+
speech_volume_normalize: null
|
| 263 |
+
rir_scp: null
|
| 264 |
+
rir_apply_prob: 1.0
|
| 265 |
+
noise_scp: null
|
| 266 |
+
noise_apply_prob: 1.0
|
| 267 |
+
noise_db_range: '13_15'
|
| 268 |
+
short_noise_thres: 0.5
|
| 269 |
+
frontend: default
|
| 270 |
+
frontend_conf:
|
| 271 |
+
n_fft: 512
|
| 272 |
+
win_length: 400
|
| 273 |
+
hop_length: 160
|
| 274 |
+
fs: 16k
|
| 275 |
+
specaug: specaug
|
| 276 |
+
specaug_conf:
|
| 277 |
+
apply_time_warp: true
|
| 278 |
+
time_warp_window: 5
|
| 279 |
+
time_warp_mode: bicubic
|
| 280 |
+
apply_freq_mask: true
|
| 281 |
+
freq_mask_width_range:
|
| 282 |
+
- 0
|
| 283 |
+
- 27
|
| 284 |
+
num_freq_mask: 2
|
| 285 |
+
apply_time_mask: true
|
| 286 |
+
time_mask_width_ratio_range:
|
| 287 |
+
- 0.0
|
| 288 |
+
- 0.05
|
| 289 |
+
num_time_mask: 2
|
| 290 |
+
normalize: global_mvn
|
| 291 |
+
normalize_conf:
|
| 292 |
+
stats_file: exp/asr_stats_raw_en_char_sp/train/feats_stats.npz
|
| 293 |
+
model: espnet
|
| 294 |
+
model_conf:
|
| 295 |
+
ctc_weight: 0.3
|
| 296 |
+
lsm_weight: 0.1
|
| 297 |
+
length_normalized_loss: false
|
| 298 |
+
preencoder: null
|
| 299 |
+
preencoder_conf: {}
|
| 300 |
+
encoder: conformer
|
| 301 |
+
encoder_conf:
|
| 302 |
+
output_size: 256
|
| 303 |
+
attention_heads: 4
|
| 304 |
+
linear_units: 2048
|
| 305 |
+
num_blocks: 12
|
| 306 |
+
dropout_rate: 0.1
|
| 307 |
+
positional_dropout_rate: 0.1
|
| 308 |
+
attention_dropout_rate: 0.1
|
| 309 |
+
input_layer: conv2d
|
| 310 |
+
normalize_before: true
|
| 311 |
+
macaron_style: true
|
| 312 |
+
rel_pos_type: latest
|
| 313 |
+
pos_enc_layer_type: rel_pos
|
| 314 |
+
selfattention_layer_type: rel_selfattn
|
| 315 |
+
activation_type: swish
|
| 316 |
+
use_cnn_module: true
|
| 317 |
+
cnn_module_kernel: 31
|
| 318 |
+
postencoder: null
|
| 319 |
+
postencoder_conf: {}
|
| 320 |
+
decoder: transformer
|
| 321 |
+
decoder_conf:
|
| 322 |
+
attention_heads: 4
|
| 323 |
+
linear_units: 2048
|
| 324 |
+
num_blocks: 6
|
| 325 |
+
dropout_rate: 0.1
|
| 326 |
+
positional_dropout_rate: 0.1
|
| 327 |
+
self_attention_dropout_rate: 0.1
|
| 328 |
+
src_attention_dropout_rate: 0.1
|
| 329 |
+
preprocessor: default
|
| 330 |
+
preprocessor_conf: {}
|
| 331 |
+
required:
|
| 332 |
+
- output_dir
|
| 333 |
+
- token_list
|
| 334 |
+
version: '202211'
|
| 335 |
+
distributed: true
|
| 336 |
+
```
|
| 337 |
+
|
| 338 |
+
</details>
|
| 339 |
+
|
| 340 |
+
|
| 341 |
+
|
| 342 |
+
### Citing ESPnet
|
| 343 |
+
|
| 344 |
+
```BibTex
|
| 345 |
+
@inproceedings{watanabe2018espnet,
|
| 346 |
+
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},
|
| 347 |
+
title={{ESPnet}: End-to-End Speech Processing Toolkit},
|
| 348 |
+
year={2018},
|
| 349 |
+
booktitle={Proceedings of Interspeech},
|
| 350 |
+
pages={2207--2211},
|
| 351 |
+
doi={10.21437/Interspeech.2018-1456},
|
| 352 |
+
url={http://dx.doi.org/10.21437/Interspeech.2018-1456}
|
| 353 |
+
}
|
| 354 |
+
|
| 355 |
+
|
| 356 |
+
|
| 357 |
+
|
| 358 |
+
```
|
| 359 |
+
|
| 360 |
+
or arXiv:
|
| 361 |
+
|
| 362 |
+
```bibtex
|
| 363 |
+
@misc{watanabe2018espnet,
|
| 364 |
+
title={ESPnet: End-to-End Speech Processing Toolkit},
|
| 365 |
+
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},
|
| 366 |
+
year={2018},
|
| 367 |
+
eprint={1804.00015},
|
| 368 |
+
archivePrefix={arXiv},
|
| 369 |
+
primaryClass={cs.CL}
|
| 370 |
+
}
|
| 371 |
+
```
|
data/nlsyms.txt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
<*IN*>
|
| 2 |
+
<*MR.*>
|
| 3 |
+
<NOISE>
|
exp/asr_stats_raw_en_char_sp/train/feats_stats.npz
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:b2aca9a9d9bf95ba2222dd91ed8d62655135d2637bf76704cf984b7d0e90756c
|
| 3 |
+
size 1402
|
exp/asr_train_asr_conformer_e12_linear2048_raw_en_char_sp/RESULTS.md
ADDED
|
@@ -0,0 +1,57 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
<!-- Generated by scripts/utils/show_asr_result.sh -->
|
| 2 |
+
# RESULTS
|
| 3 |
+
## Environments
|
| 4 |
+
- date: `Wed Dec 28 20:41:40 EST 2022`
|
| 5 |
+
- python version: `3.9.15 (main, Nov 24 2022, 14:31:59) [GCC 11.2.0]`
|
| 6 |
+
- espnet version: `espnet 202211`
|
| 7 |
+
- pytorch version: `pytorch 1.12.1`
|
| 8 |
+
- Git hash: `ad91279f0108d54bd22abe29671b376f048822c5`
|
| 9 |
+
- Commit date: `Wed Dec 28 20:15:42 2022 -0500`
|
| 10 |
+
|
| 11 |
+
## asr_train_asr_conformer_e12_linear2048_raw_en_char_sp
|
| 12 |
+
### WER
|
| 13 |
+
|
| 14 |
+
|dataset|Snt|Wrd|Corr|Sub|Del|Ins|Err|S.Err|
|
| 15 |
+
|---|---|---|---|---|---|---|---|---|
|
| 16 |
+
|decode_asr_lm_lm_train_lm_en_char_valid.loss.ave_asr_model_valid.acc.ave/dt05_real_beamformit_2mics|1640|27119|88.8|9.4|1.8|0.9|12.2|74.8|
|
| 17 |
+
|decode_asr_lm_lm_train_lm_en_char_valid.loss.ave_asr_model_valid.acc.ave/dt05_real_beamformit_5mics|1640|27119|90.2|8.1|1.6|0.8|10.5|71.6|
|
| 18 |
+
|decode_asr_lm_lm_train_lm_en_char_valid.loss.ave_asr_model_valid.acc.ave/dt05_real_isolated_1ch_track|1640|27119|87.4|10.5|2.1|1.1|13.7|77.0|
|
| 19 |
+
|decode_asr_lm_lm_train_lm_en_char_valid.loss.ave_asr_model_valid.acc.ave/dt05_simu_beamformit_2mics|1640|27120|86.4|11.1|2.4|1.3|14.8|78.9|
|
| 20 |
+
|decode_asr_lm_lm_train_lm_en_char_valid.loss.ave_asr_model_valid.acc.ave/dt05_simu_beamformit_5mics|1640|27120|88.6|9.4|2.0|1.1|12.5|75.4|
|
| 21 |
+
|decode_asr_lm_lm_train_lm_en_char_valid.loss.ave_asr_model_valid.acc.ave/dt05_simu_isolated_1ch_track|1640|27120|85.1|12.4|2.5|1.4|16.3|79.9|
|
| 22 |
+
|decode_asr_lm_lm_train_lm_en_char_valid.loss.ave_asr_model_valid.acc.ave/et05_real_beamformit_2mics|1320|21409|82.7|14.6|2.7|1.6|18.9|83.1|
|
| 23 |
+
|decode_asr_lm_lm_train_lm_en_char_valid.loss.ave_asr_model_valid.acc.ave/et05_real_beamformit_5mics|1320|21409|85.2|12.4|2.4|1.5|16.3|79.5|
|
| 24 |
+
|decode_asr_lm_lm_train_lm_en_char_valid.loss.ave_asr_model_valid.acc.ave/et05_real_isolated_1ch_track|1320|21409|80.8|16.2|3.1|2.1|21.3|85.2|
|
| 25 |
+
|decode_asr_lm_lm_train_lm_en_char_valid.loss.ave_asr_model_valid.acc.ave/et05_simu_beamformit_2mics|1320|21416|81.1|15.2|3.7|1.9|20.8|84.8|
|
| 26 |
+
|decode_asr_lm_lm_train_lm_en_char_valid.loss.ave_asr_model_valid.acc.ave/et05_simu_beamformit_5mics|1320|21416|83.6|13.4|3.0|1.7|18.1|81.4|
|
| 27 |
+
|decode_asr_lm_lm_train_lm_en_char_valid.loss.ave_asr_model_valid.acc.ave/et05_simu_isolated_1ch_track|1320|21416|79.2|17.3|3.5|2.4|23.1|83.4|
|
| 28 |
+
|decode_asr_lm_lm_train_lm_transformer_en_char_valid.loss.ave_asr_model_valid.acc.ave/dt05_real_beamformit_5mics|1640|27119|93.3|5.4|1.3|0.5|7.3|55.6|
|
| 29 |
+
|decode_asr_lm_lm_train_lm_transformer_en_char_valid.loss.ave_asr_model_valid.acc.ave/dt05_simu_beamformit_5mics|1640|27120|91.7|6.7|1.6|0.9|9.1|62.0|
|
| 30 |
+
|decode_asr_lm_lm_train_lm_transformer_en_char_valid.loss.ave_asr_model_valid.acc.ave/et05_real_beamformit_5mics|1320|21409|89.2|8.9|1.9|1.1|12.0|64.5|
|
| 31 |
+
|decode_asr_lm_lm_train_lm_transformer_en_char_valid.loss.ave_asr_model_valid.acc.ave/et05_simu_beamformit_5mics|1320|21416|87.8|9.6|2.6|1.4|13.6|68.1|
|
| 32 |
+
|
| 33 |
+
### CER
|
| 34 |
+
|
| 35 |
+
|dataset|Snt|Wrd|Corr|Sub|Del|Ins|Err|S.Err|
|
| 36 |
+
|---|---|---|---|---|---|---|---|---|
|
| 37 |
+
|decode_asr_lm_lm_train_lm_en_char_valid.loss.ave_asr_model_valid.acc.ave/dt05_real_beamformit_2mics|1640|160390|95.5|2.5|2.1|1.2|5.7|74.8|
|
| 38 |
+
|decode_asr_lm_lm_train_lm_en_char_valid.loss.ave_asr_model_valid.acc.ave/dt05_real_beamformit_5mics|1640|160390|96.2|2.0|1.8|0.9|4.7|71.6|
|
| 39 |
+
|decode_asr_lm_lm_train_lm_en_char_valid.loss.ave_asr_model_valid.acc.ave/dt05_real_isolated_1ch_track|1640|160390|94.8|2.9|2.3|1.3|6.5|77.0|
|
| 40 |
+
|decode_asr_lm_lm_train_lm_en_char_valid.loss.ave_asr_model_valid.acc.ave/dt05_simu_beamformit_2mics|1640|160400|94.1|3.1|2.7|1.6|7.5|78.9|
|
| 41 |
+
|decode_asr_lm_lm_train_lm_en_char_valid.loss.ave_asr_model_valid.acc.ave/dt05_simu_beamformit_5mics|1640|160400|95.3|2.5|2.2|1.3|6.0|75.4|
|
| 42 |
+
|decode_asr_lm_lm_train_lm_en_char_valid.loss.ave_asr_model_valid.acc.ave/dt05_simu_isolated_1ch_track|1640|160400|93.4|3.7|2.9|1.8|8.3|79.9|
|
| 43 |
+
|decode_asr_lm_lm_train_lm_en_char_valid.loss.ave_asr_model_valid.acc.ave/et05_real_beamformit_2mics|1320|126796|92.5|4.3|3.2|1.9|9.4|83.1|
|
| 44 |
+
|decode_asr_lm_lm_train_lm_en_char_valid.loss.ave_asr_model_valid.acc.ave/et05_real_beamformit_5mics|1320|126796|93.9|3.4|2.7|1.6|7.7|79.5|
|
| 45 |
+
|decode_asr_lm_lm_train_lm_en_char_valid.loss.ave_asr_model_valid.acc.ave/et05_real_isolated_1ch_track|1320|126796|91.3|5.1|3.6|2.3|11.0|85.2|
|
| 46 |
+
|decode_asr_lm_lm_train_lm_en_char_valid.loss.ave_asr_model_valid.acc.ave/et05_simu_beamformit_2mics|1320|126812|90.9|4.8|4.3|2.3|11.4|84.8|
|
| 47 |
+
|decode_asr_lm_lm_train_lm_en_char_valid.loss.ave_asr_model_valid.acc.ave/et05_simu_beamformit_5mics|1320|126812|92.6|3.9|3.5|2.0|9.4|81.4|
|
| 48 |
+
|decode_asr_lm_lm_train_lm_en_char_valid.loss.ave_asr_model_valid.acc.ave/et05_simu_isolated_1ch_track|1320|126812|89.9|5.7|4.3|2.9|12.9|83.4|
|
| 49 |
+
|decode_asr_lm_lm_train_lm_transformer_en_char_valid.loss.ave_asr_model_valid.acc.ave/dt05_real_beamformit_5mics|1640|160390|97.2|1.5|1.3|0.7|3.5|55.6|
|
| 50 |
+
|decode_asr_lm_lm_train_lm_transformer_en_char_valid.loss.ave_asr_model_valid.acc.ave/dt05_simu_beamformit_5mics|1640|160400|96.3|2.0|1.7|1.0|4.7|62.0|
|
| 51 |
+
|decode_asr_lm_lm_train_lm_transformer_en_char_valid.loss.ave_asr_model_valid.acc.ave/et05_real_beamformit_5mics|1320|126796|95.1|2.8|2.1|1.2|6.1|64.6|
|
| 52 |
+
|decode_asr_lm_lm_train_lm_transformer_en_char_valid.loss.ave_asr_model_valid.acc.ave/et05_simu_beamformit_5mics|1320|126812|94.0|3.1|3.0|1.6|7.7|68.1|
|
| 53 |
+
|
| 54 |
+
### TER
|
| 55 |
+
|
| 56 |
+
|dataset|Snt|Wrd|Corr|Sub|Del|Ins|Err|S.Err|
|
| 57 |
+
|---|---|---|---|---|---|---|---|---|
|
exp/asr_train_asr_conformer_e12_linear2048_raw_en_char_sp/config.yaml
ADDED
|
@@ -0,0 +1,266 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
config: conf/tuning/train_asr_conformer_e12_linear2048.yaml
|
| 2 |
+
print_config: false
|
| 3 |
+
log_level: INFO
|
| 4 |
+
dry_run: false
|
| 5 |
+
iterator_type: sequence
|
| 6 |
+
output_dir: exp/asr_train_asr_conformer_e12_linear2048_raw_en_char_sp
|
| 7 |
+
ngpu: 1
|
| 8 |
+
seed: 2022
|
| 9 |
+
num_workers: 4
|
| 10 |
+
num_att_plot: 3
|
| 11 |
+
dist_backend: nccl
|
| 12 |
+
dist_init_method: env://
|
| 13 |
+
dist_world_size: 2
|
| 14 |
+
dist_rank: 0
|
| 15 |
+
local_rank: 0
|
| 16 |
+
dist_master_addr: localhost
|
| 17 |
+
dist_master_port: 45069
|
| 18 |
+
dist_launcher: null
|
| 19 |
+
multiprocessing_distributed: true
|
| 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: 50
|
| 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 |
+
- acc
|
| 39 |
+
- max
|
| 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: 1
|
| 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: 15000000
|
| 69 |
+
valid_batch_bins: null
|
| 70 |
+
train_shape_file:
|
| 71 |
+
- exp/asr_stats_raw_en_char_sp/train/speech_shape
|
| 72 |
+
- exp/asr_stats_raw_en_char_sp/train/text_shape.char
|
| 73 |
+
valid_shape_file:
|
| 74 |
+
- exp/asr_stats_raw_en_char_sp/valid/speech_shape
|
| 75 |
+
- exp/asr_stats_raw_en_char_sp/valid/text_shape.char
|
| 76 |
+
batch_type: numel
|
| 77 |
+
valid_batch_type: null
|
| 78 |
+
fold_length:
|
| 79 |
+
- 80000
|
| 80 |
+
- 150
|
| 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/tr05_multi_noisy_si284_sp/wav.scp
|
| 89 |
+
- speech
|
| 90 |
+
- kaldi_ark
|
| 91 |
+
- - dump/raw/tr05_multi_noisy_si284_sp/text
|
| 92 |
+
- text
|
| 93 |
+
- text
|
| 94 |
+
valid_data_path_and_name_and_type:
|
| 95 |
+
- - dump/raw/dt05_multi_isolated_1ch_track/wav.scp
|
| 96 |
+
- speech
|
| 97 |
+
- kaldi_ark
|
| 98 |
+
- - dump/raw/dt05_multi_isolated_1ch_track/text
|
| 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: 0.001
|
| 108 |
+
weight_decay: 1.0e-06
|
| 109 |
+
scheduler: warmuplr
|
| 110 |
+
scheduler_conf:
|
| 111 |
+
warmup_steps: 25000
|
| 112 |
+
token_list:
|
| 113 |
+
- <blank>
|
| 114 |
+
- <unk>
|
| 115 |
+
- <space>
|
| 116 |
+
- E
|
| 117 |
+
- T
|
| 118 |
+
- A
|
| 119 |
+
- N
|
| 120 |
+
- I
|
| 121 |
+
- O
|
| 122 |
+
- S
|
| 123 |
+
- R
|
| 124 |
+
- H
|
| 125 |
+
- L
|
| 126 |
+
- D
|
| 127 |
+
- C
|
| 128 |
+
- U
|
| 129 |
+
- M
|
| 130 |
+
- P
|
| 131 |
+
- F
|
| 132 |
+
- G
|
| 133 |
+
- Y
|
| 134 |
+
- W
|
| 135 |
+
- B
|
| 136 |
+
- V
|
| 137 |
+
- K
|
| 138 |
+
- .
|
| 139 |
+
- X
|
| 140 |
+
- ''''
|
| 141 |
+
- J
|
| 142 |
+
- Q
|
| 143 |
+
- Z
|
| 144 |
+
- ','
|
| 145 |
+
- '-'
|
| 146 |
+
- '"'
|
| 147 |
+
- <NOISE>
|
| 148 |
+
- '*'
|
| 149 |
+
- ':'
|
| 150 |
+
- (
|
| 151 |
+
- )
|
| 152 |
+
- '?'
|
| 153 |
+
- '&'
|
| 154 |
+
- ;
|
| 155 |
+
- '!'
|
| 156 |
+
- /
|
| 157 |
+
- '{'
|
| 158 |
+
- '}'
|
| 159 |
+
- '1'
|
| 160 |
+
- '2'
|
| 161 |
+
- '0'
|
| 162 |
+
- $
|
| 163 |
+
- '8'
|
| 164 |
+
- '9'
|
| 165 |
+
- '6'
|
| 166 |
+
- '3'
|
| 167 |
+
- '5'
|
| 168 |
+
- '7'
|
| 169 |
+
- '4'
|
| 170 |
+
- '~'
|
| 171 |
+
- '`'
|
| 172 |
+
- _
|
| 173 |
+
- <*IN*>
|
| 174 |
+
- <*MR.*>
|
| 175 |
+
- \
|
| 176 |
+
- ^
|
| 177 |
+
- <sos/eos>
|
| 178 |
+
init: null
|
| 179 |
+
input_size: null
|
| 180 |
+
ctc_conf:
|
| 181 |
+
dropout_rate: 0.0
|
| 182 |
+
ctc_type: builtin
|
| 183 |
+
reduce: true
|
| 184 |
+
ignore_nan_grad: null
|
| 185 |
+
zero_infinity: true
|
| 186 |
+
joint_net_conf: null
|
| 187 |
+
use_preprocessor: true
|
| 188 |
+
token_type: char
|
| 189 |
+
bpemodel: null
|
| 190 |
+
non_linguistic_symbols: data/nlsyms.txt
|
| 191 |
+
cleaner: null
|
| 192 |
+
g2p: null
|
| 193 |
+
speech_volume_normalize: null
|
| 194 |
+
rir_scp: null
|
| 195 |
+
rir_apply_prob: 1.0
|
| 196 |
+
noise_scp: null
|
| 197 |
+
noise_apply_prob: 1.0
|
| 198 |
+
noise_db_range: '13_15'
|
| 199 |
+
short_noise_thres: 0.5
|
| 200 |
+
frontend: default
|
| 201 |
+
frontend_conf:
|
| 202 |
+
n_fft: 512
|
| 203 |
+
win_length: 400
|
| 204 |
+
hop_length: 160
|
| 205 |
+
fs: 16k
|
| 206 |
+
specaug: specaug
|
| 207 |
+
specaug_conf:
|
| 208 |
+
apply_time_warp: true
|
| 209 |
+
time_warp_window: 5
|
| 210 |
+
time_warp_mode: bicubic
|
| 211 |
+
apply_freq_mask: true
|
| 212 |
+
freq_mask_width_range:
|
| 213 |
+
- 0
|
| 214 |
+
- 27
|
| 215 |
+
num_freq_mask: 2
|
| 216 |
+
apply_time_mask: true
|
| 217 |
+
time_mask_width_ratio_range:
|
| 218 |
+
- 0.0
|
| 219 |
+
- 0.05
|
| 220 |
+
num_time_mask: 2
|
| 221 |
+
normalize: global_mvn
|
| 222 |
+
normalize_conf:
|
| 223 |
+
stats_file: exp/asr_stats_raw_en_char_sp/train/feats_stats.npz
|
| 224 |
+
model: espnet
|
| 225 |
+
model_conf:
|
| 226 |
+
ctc_weight: 0.3
|
| 227 |
+
lsm_weight: 0.1
|
| 228 |
+
length_normalized_loss: false
|
| 229 |
+
preencoder: null
|
| 230 |
+
preencoder_conf: {}
|
| 231 |
+
encoder: conformer
|
| 232 |
+
encoder_conf:
|
| 233 |
+
output_size: 256
|
| 234 |
+
attention_heads: 4
|
| 235 |
+
linear_units: 2048
|
| 236 |
+
num_blocks: 12
|
| 237 |
+
dropout_rate: 0.1
|
| 238 |
+
positional_dropout_rate: 0.1
|
| 239 |
+
attention_dropout_rate: 0.1
|
| 240 |
+
input_layer: conv2d
|
| 241 |
+
normalize_before: true
|
| 242 |
+
macaron_style: true
|
| 243 |
+
rel_pos_type: latest
|
| 244 |
+
pos_enc_layer_type: rel_pos
|
| 245 |
+
selfattention_layer_type: rel_selfattn
|
| 246 |
+
activation_type: swish
|
| 247 |
+
use_cnn_module: true
|
| 248 |
+
cnn_module_kernel: 31
|
| 249 |
+
postencoder: null
|
| 250 |
+
postencoder_conf: {}
|
| 251 |
+
decoder: transformer
|
| 252 |
+
decoder_conf:
|
| 253 |
+
attention_heads: 4
|
| 254 |
+
linear_units: 2048
|
| 255 |
+
num_blocks: 6
|
| 256 |
+
dropout_rate: 0.1
|
| 257 |
+
positional_dropout_rate: 0.1
|
| 258 |
+
self_attention_dropout_rate: 0.1
|
| 259 |
+
src_attention_dropout_rate: 0.1
|
| 260 |
+
preprocessor: default
|
| 261 |
+
preprocessor_conf: {}
|
| 262 |
+
required:
|
| 263 |
+
- output_dir
|
| 264 |
+
- token_list
|
| 265 |
+
version: '202211'
|
| 266 |
+
distributed: true
|
exp/asr_train_asr_conformer_e12_linear2048_raw_en_char_sp/images/acc.png
ADDED
|
exp/asr_train_asr_conformer_e12_linear2048_raw_en_char_sp/images/backward_time.png
ADDED
|
exp/asr_train_asr_conformer_e12_linear2048_raw_en_char_sp/images/cer.png
ADDED
|
exp/asr_train_asr_conformer_e12_linear2048_raw_en_char_sp/images/cer_ctc.png
ADDED
|
exp/asr_train_asr_conformer_e12_linear2048_raw_en_char_sp/images/forward_time.png
ADDED
|
exp/asr_train_asr_conformer_e12_linear2048_raw_en_char_sp/images/gpu_max_cached_mem_GB.png
ADDED
|
exp/asr_train_asr_conformer_e12_linear2048_raw_en_char_sp/images/iter_time.png
ADDED
|
exp/asr_train_asr_conformer_e12_linear2048_raw_en_char_sp/images/loss.png
ADDED
|
exp/asr_train_asr_conformer_e12_linear2048_raw_en_char_sp/images/loss_att.png
ADDED
|
exp/asr_train_asr_conformer_e12_linear2048_raw_en_char_sp/images/loss_ctc.png
ADDED
|
exp/asr_train_asr_conformer_e12_linear2048_raw_en_char_sp/images/optim0_lr0.png
ADDED
|
exp/asr_train_asr_conformer_e12_linear2048_raw_en_char_sp/images/optim_step_time.png
ADDED
|
exp/asr_train_asr_conformer_e12_linear2048_raw_en_char_sp/images/train_time.png
ADDED
|
exp/asr_train_asr_conformer_e12_linear2048_raw_en_char_sp/images/wer.png
ADDED
|
exp/asr_train_asr_conformer_e12_linear2048_raw_en_char_sp/valid.acc.ave_10best.pth
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:899f69f8918373c07bbbf5286c424363f3ba92ce84c3f9ecf7226e2f7025bf3c
|
| 3 |
+
size 172493605
|
exp/lm_train_lm_transformer_en_char/12epoch.pth
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:13dc5af025af5be3560922587402820207a78d258eb78fe1d47553a62b4b5895
|
| 3 |
+
size 202293743
|
exp/lm_train_lm_transformer_en_char/config.yaml
ADDED
|
@@ -0,0 +1,189 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
accum_grad: 4
|
| 2 |
+
allow_variable_data_keys: false
|
| 3 |
+
batch_bins: 150000
|
| 4 |
+
batch_size: 20
|
| 5 |
+
batch_type: numel
|
| 6 |
+
best_model_criterion:
|
| 7 |
+
- - valid
|
| 8 |
+
- loss
|
| 9 |
+
- min
|
| 10 |
+
bpemodel: null
|
| 11 |
+
chunk_length: 500
|
| 12 |
+
chunk_shift_ratio: 0.5
|
| 13 |
+
cleaner: null
|
| 14 |
+
collect_stats: false
|
| 15 |
+
config: conf/train_lm_transformer.yaml
|
| 16 |
+
cudnn_benchmark: false
|
| 17 |
+
cudnn_deterministic: true
|
| 18 |
+
cudnn_enabled: true
|
| 19 |
+
detect_anomaly: false
|
| 20 |
+
dist_backend: nccl
|
| 21 |
+
dist_init_method: env://
|
| 22 |
+
dist_launcher: null
|
| 23 |
+
dist_master_addr: null
|
| 24 |
+
dist_master_port: null
|
| 25 |
+
dist_rank: null
|
| 26 |
+
dist_world_size: null
|
| 27 |
+
distributed: false
|
| 28 |
+
dry_run: false
|
| 29 |
+
early_stopping_criterion:
|
| 30 |
+
- valid
|
| 31 |
+
- loss
|
| 32 |
+
- min
|
| 33 |
+
fold_length:
|
| 34 |
+
- 150
|
| 35 |
+
freeze_param: []
|
| 36 |
+
g2p: null
|
| 37 |
+
grad_clip: 5.0
|
| 38 |
+
grad_clip_type: 2.0
|
| 39 |
+
grad_noise: false
|
| 40 |
+
ignore_init_mismatch: false
|
| 41 |
+
init: null
|
| 42 |
+
init_param: []
|
| 43 |
+
iterator_type: sequence
|
| 44 |
+
keep_nbest_models: 10
|
| 45 |
+
lm: transformer
|
| 46 |
+
lm_conf:
|
| 47 |
+
att_unit: 512
|
| 48 |
+
dropout_rate: 0.1
|
| 49 |
+
embed_unit: 128
|
| 50 |
+
head: 8
|
| 51 |
+
layer: 16
|
| 52 |
+
pos_enc: null
|
| 53 |
+
unit: 2048
|
| 54 |
+
local_rank: 0
|
| 55 |
+
log_interval: null
|
| 56 |
+
log_level: INFO
|
| 57 |
+
max_cache_fd: 32
|
| 58 |
+
max_cache_size: 0.0
|
| 59 |
+
max_epoch: 30
|
| 60 |
+
model_conf:
|
| 61 |
+
ignore_id: 0
|
| 62 |
+
multiple_iterator: false
|
| 63 |
+
multiprocessing_distributed: false
|
| 64 |
+
nbest_averaging_interval: 0
|
| 65 |
+
ngpu: 1
|
| 66 |
+
no_forward_run: false
|
| 67 |
+
non_linguistic_symbols: data/nlsyms.txt
|
| 68 |
+
num_att_plot: 3
|
| 69 |
+
num_cache_chunks: 1024
|
| 70 |
+
num_iters_per_epoch: null
|
| 71 |
+
num_workers: 1
|
| 72 |
+
optim: adam
|
| 73 |
+
optim_conf:
|
| 74 |
+
lr: 0.001
|
| 75 |
+
output_dir: exp_real2/lm_train_lm_transformer_en_char
|
| 76 |
+
patience: null
|
| 77 |
+
pretrain_path: null
|
| 78 |
+
print_config: false
|
| 79 |
+
required:
|
| 80 |
+
- output_dir
|
| 81 |
+
- token_list
|
| 82 |
+
resume: true
|
| 83 |
+
scheduler: warmuplr
|
| 84 |
+
scheduler_conf:
|
| 85 |
+
warmup_steps: 25000
|
| 86 |
+
seed: 0
|
| 87 |
+
sharded_ddp: false
|
| 88 |
+
sort_batch: descending
|
| 89 |
+
sort_in_batch: descending
|
| 90 |
+
token_list:
|
| 91 |
+
- <blank>
|
| 92 |
+
- <unk>
|
| 93 |
+
- <space>
|
| 94 |
+
- E
|
| 95 |
+
- T
|
| 96 |
+
- A
|
| 97 |
+
- N
|
| 98 |
+
- I
|
| 99 |
+
- O
|
| 100 |
+
- S
|
| 101 |
+
- R
|
| 102 |
+
- H
|
| 103 |
+
- L
|
| 104 |
+
- D
|
| 105 |
+
- C
|
| 106 |
+
- U
|
| 107 |
+
- M
|
| 108 |
+
- P
|
| 109 |
+
- F
|
| 110 |
+
- G
|
| 111 |
+
- Y
|
| 112 |
+
- W
|
| 113 |
+
- B
|
| 114 |
+
- V
|
| 115 |
+
- K
|
| 116 |
+
- .
|
| 117 |
+
- X
|
| 118 |
+
- ''''
|
| 119 |
+
- J
|
| 120 |
+
- Q
|
| 121 |
+
- Z
|
| 122 |
+
- ','
|
| 123 |
+
- '-'
|
| 124 |
+
- '"'
|
| 125 |
+
- <NOISE>
|
| 126 |
+
- '*'
|
| 127 |
+
- ':'
|
| 128 |
+
- (
|
| 129 |
+
- )
|
| 130 |
+
- '?'
|
| 131 |
+
- '&'
|
| 132 |
+
- ;
|
| 133 |
+
- '!'
|
| 134 |
+
- /
|
| 135 |
+
- '{'
|
| 136 |
+
- '}'
|
| 137 |
+
- '1'
|
| 138 |
+
- '2'
|
| 139 |
+
- '0'
|
| 140 |
+
- $
|
| 141 |
+
- '8'
|
| 142 |
+
- '9'
|
| 143 |
+
- '6'
|
| 144 |
+
- '3'
|
| 145 |
+
- '5'
|
| 146 |
+
- '7'
|
| 147 |
+
- '4'
|
| 148 |
+
- '~'
|
| 149 |
+
- '`'
|
| 150 |
+
- _
|
| 151 |
+
- <*IN*>
|
| 152 |
+
- <*MR.*>
|
| 153 |
+
- \
|
| 154 |
+
- ^
|
| 155 |
+
- <sos/eos>
|
| 156 |
+
token_type: char
|
| 157 |
+
train_data_path_and_name_and_type:
|
| 158 |
+
- - dump/raw/lm_train.txt
|
| 159 |
+
- text
|
| 160 |
+
- text
|
| 161 |
+
train_dtype: float32
|
| 162 |
+
train_shape_file:
|
| 163 |
+
- exp_real2/lm_stats_en_char/train/text_shape.char
|
| 164 |
+
unused_parameters: false
|
| 165 |
+
use_amp: false
|
| 166 |
+
use_matplotlib: true
|
| 167 |
+
use_preprocessor: true
|
| 168 |
+
use_tensorboard: true
|
| 169 |
+
use_wandb: false
|
| 170 |
+
val_scheduler_criterion:
|
| 171 |
+
- valid
|
| 172 |
+
- loss
|
| 173 |
+
valid_batch_bins: null
|
| 174 |
+
valid_batch_size: null
|
| 175 |
+
valid_batch_type: null
|
| 176 |
+
valid_data_path_and_name_and_type:
|
| 177 |
+
- - dump/raw/dt05_multi_isolated_1ch_track/text
|
| 178 |
+
- text
|
| 179 |
+
- text
|
| 180 |
+
valid_max_cache_size: null
|
| 181 |
+
valid_shape_file:
|
| 182 |
+
- exp_real2/lm_stats_en_char/valid/text_shape.char
|
| 183 |
+
version: 0.10.7a1
|
| 184 |
+
wandb_entity: null
|
| 185 |
+
wandb_id: null
|
| 186 |
+
wandb_model_log_interval: -1
|
| 187 |
+
wandb_name: null
|
| 188 |
+
wandb_project: null
|
| 189 |
+
write_collected_feats: false
|
exp/lm_train_lm_transformer_en_char/images/backward_time.png
ADDED
|
exp/lm_train_lm_transformer_en_char/images/forward_time.png
ADDED
|
exp/lm_train_lm_transformer_en_char/images/gpu_max_cached_mem_GB.png
ADDED
|
exp/lm_train_lm_transformer_en_char/images/iter_time.png
ADDED
|
exp/lm_train_lm_transformer_en_char/images/loss.png
ADDED
|
exp/lm_train_lm_transformer_en_char/images/optim0_lr0.png
ADDED
|
exp/lm_train_lm_transformer_en_char/images/optim_step_time.png
ADDED
|
exp/lm_train_lm_transformer_en_char/images/train_time.png
ADDED
|
exp/lm_train_lm_transformer_en_char/perplexity_test/ppl
ADDED
|
@@ -0,0 +1 @@
|
|
|
|
|
|
|
| 1 |
+
1.8016810278173603
|
meta.yaml
ADDED
|
@@ -0,0 +1,10 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
espnet: '202211'
|
| 2 |
+
files:
|
| 3 |
+
asr_model_file: exp/asr_train_asr_conformer_e12_linear2048_raw_en_char_sp/valid.acc.ave_10best.pth
|
| 4 |
+
lm_file: exp/lm_train_lm_transformer_en_char/12epoch.pth
|
| 5 |
+
python: "3.9.15 (main, Nov 24 2022, 14:31:59) \n[GCC 11.2.0]"
|
| 6 |
+
timestamp: 1672282123.487957
|
| 7 |
+
torch: 1.12.1
|
| 8 |
+
yaml_files:
|
| 9 |
+
asr_train_config: exp/asr_train_asr_conformer_e12_linear2048_raw_en_char_sp/config.yaml
|
| 10 |
+
lm_train_config: exp/lm_train_lm_transformer_en_char/config.yaml
|