|
downstream_expert: |
|
datarc: |
|
batch_size: 32 |
|
bucket_file: /home/leo/d/datasets/LibriSpeech/len_for_bucket |
|
decoder_args: |
|
beam: 5 |
|
beam_threshold: 25 |
|
criterion: ctc |
|
decoder_type: None |
|
kenlm_model: /path/to/KenLM |
|
lexicon: /path/to/4-gram.arpa |
|
lm_weight: 2 |
|
nbest: 1 |
|
sil_weight: 0 |
|
unk_weight: -math.inf |
|
word_score: -1 |
|
dev-clean: |
|
- dev-clean |
|
dev-other: |
|
- dev-other |
|
dict_path: ./downstream/asr/char.dict |
|
eval_batch_size: 1 |
|
libri_root: /home/leo/d/datasets/LibriSpeech |
|
num_workers: 12 |
|
test-clean: |
|
- test-clean |
|
test-other: |
|
- test-other |
|
train: |
|
- train-clean-100 |
|
train_batch_size: 32 |
|
zero_infinity: true |
|
modelrc: |
|
RNNs: |
|
bidirection: true |
|
dim: |
|
- 1024 |
|
- 1024 |
|
dropout: |
|
- 0.2 |
|
- 0.2 |
|
layer_norm: |
|
- false |
|
- false |
|
module: LSTM |
|
proj: |
|
- false |
|
- false |
|
sample_rate: |
|
- 1 |
|
- 1 |
|
sample_style: concat |
|
total_rate: -1 |
|
Wav2Letter: |
|
total_rate: 320 |
|
project_dim: 1024 |
|
select: RNNs |
|
optimizer: |
|
lr: 0.0001 |
|
name: TorchOptim |
|
torch_optim_name: Adam |
|
runner: |
|
eval_dataloaders: |
|
- dev-clean |
|
eval_step: 2000 |
|
gradient_accumulate_steps: 1 |
|
gradient_clipping: 1 |
|
log_step: 100 |
|
max_keep: 1 |
|
save_step: 5 |
|
total_steps: 10 |
|
specaug: |
|
apply_freq_mask: true |
|
apply_time_mask: true |
|
apply_time_warp: true |
|
freq_mask_width_range: |
|
- 0 |
|
- 50 |
|
num_freq_mask: 4 |
|
num_time_mask: 2 |
|
time_mask_width_range: |
|
- 0 |
|
- 40 |
|
time_warp_window: 5 |
|
|