| # minibatch related | |
| # # dataset_type: small | |
| batch_type: length | |
| batch_bins: 2000 | |
| num_workers: 16 | |
| speech_length_min: 100 | |
| speech_length_max: 15000 | |
| # dataset_type: large | |
| dataset_conf: | |
| data_types: sound,text,text | |
| filter_conf: | |
| speech_length_min: 100 | |
| speech_length_max: 15000 | |
| token_length_min: 0 | |
| token_length_max: 200 | |
| shuffle: true | |
| shuffle_conf: | |
| shuffle_size: 2048 | |
| sort_size: 500 | |
| batch_conf: | |
| batch_type: 'token' | |
| batch_size: 120000 | |
| num_workers: 16 | |
| # optimization related | |
| accum_grad: 1 | |
| grad_clip: 5 | |
| max_epoch: 40 | |
| keep_nbest_models: 10 | |
| optim: adam | |
| optim_conf: | |
| lr: 0.0005 | |
| scheduler: warmuplr | |
| scheduler_conf: | |
| warmup_steps: 30000 | |
| log_interval: 50 | |