| File was renamed from egs/alimeeting/sa-asr/conf/train_asr_conformer.yaml |
| | |
|---|
| | | # network architecture |
|---|
| | | frontend: default |
|---|
| | | frontend: multichannelfrontend |
|---|
| | | frontend_conf: |
|---|
| | | fs: 16000 |
|---|
| | | window: hann |
|---|
| | | n_fft: 400 |
|---|
| | | win_length: 400 |
|---|
| | | hop_length: 160 |
|---|
| | | n_mels: 80 |
|---|
| | | frame_length: 25 |
|---|
| | | frame_shift: 10 |
|---|
| | | lfr_m: 1 |
|---|
| | | lfr_n: 1 |
|---|
| | | use_channel: 0 |
|---|
| | | |
|---|
| | | # encoder related |
|---|
| | |
|---|
| | | lsm_weight: 0.1 # label smoothing option |
|---|
| | | length_normalized_loss: false |
|---|
| | | |
|---|
| | | # minibatch related |
|---|
| | | batch_type: numel |
|---|
| | | batch_bins: 10000000 # reduce/increase this number according to your GPU memory |
|---|
| | | |
|---|
| | | dataset_conf: |
|---|
| | | data_names: speech,text |
|---|
| | | data_types: sound,text |
|---|
| | | shuffle: True |
|---|
| | | shuffle_conf: |
|---|
| | | shuffle_size: 2048 |
|---|
| | | sort_size: 500 |
|---|
| | | batch_conf: |
|---|
| | | batch_type: token |
|---|
| | | batch_size: 7000 |
|---|
| | | num_workers: 8 |
|---|
| | | |
|---|
| | | # optimization related |
|---|
| | | accum_grad: 1 |
|---|
