| | |
|---|
| | | # to print the register_table: |
|---|
| | | # from funasr.register import tables |
|---|
| | | # tables.print() |
|---|
| | | |
|---|
| | | # network architecture |
|---|
| | | model: SenseVoice |
|---|
| | | model_conf: |
|---|
| | | lsm_weight: 0.1 |
|---|
| | | length_normalized_loss: true |
|---|
| | | hub: funasr |
|---|
| | | activation_checkpoint: true |
|---|
| | | sos: "<|startoftranscript|>" |
|---|
| | | eos: "<|endoftext|>" |
|---|
| | | downsample_rate: 4 |
|---|
| | | use_padmask: true |
|---|
| | | |
|---|
| | | |
|---|
| | | |
|---|
| | | # only use for hub == funasr, |
|---|
| | | # if hub == openai, dims is automaticall download |
|---|
| | | dims: |
|---|
| | | n_mels: 128 |
|---|
| | | n_vocab: 51866 |
|---|
| | | n_audio_ctx: 1500 |
|---|
| | | n_audio_state: 1280 |
|---|
| | | n_audio_head: 20 |
|---|
| | | n_audio_layer: 32 |
|---|
| | | n_text_ctx: 448 |
|---|
| | | n_text_state: 1280 |
|---|
| | | n_text_head: 20 |
|---|
| | | n_text_layer: 32 |
|---|
| | | dims: |
|---|
| | | n_mels: 128 |
|---|
| | | n_vocab: 60515 |
|---|
| | | n_audio_ctx: 1500 |
|---|
| | | n_audio_state: 1280 |
|---|
| | | n_audio_head: 20 |
|---|
| | | n_audio_layer: 32 |
|---|
| | | n_text_ctx: 448 |
|---|
| | | n_text_state: 1280 |
|---|
| | | n_text_head: 20 |
|---|
| | | n_text_layer: 32 |
|---|
| | | |
|---|
| | | # frontend related |
|---|
| | | frontend: WhisperFrontend |
|---|
| | | frontend_conf: |
|---|
| | | fs: 16000 |
|---|
| | | n_mels: ${dims.n_mels} |
|---|
| | | do_pad_trim: true |
|---|
| | | n_mels: ${model_conf.dims.n_mels} |
|---|
| | | do_pad_trim: false |
|---|
| | | |
|---|
| | | tokenizer: WhisperTokenizer |
|---|
| | | tokenizer: SenseVoiceTokenizer |
|---|
| | | tokenizer_conf: |
|---|
| | | language: null |
|---|
| | | task: transcribe |
|---|
| | | vocab_path: null |
|---|
| | | is_multilingual: true |
|---|
| | | num_languages: 100 |
|---|
| | | num_languages: 8749 |
|---|
| | | |
|---|
| | | scope_map: [none, "model."] |
|---|
| | | dataset: SenseVoiceDataset |
|---|
| | | dataset_conf: |
|---|
| | | index_ds: IndexDSJsonl |
|---|
| | | batch_sampler: EspnetStyleBatchSampler |
|---|
| | | batch_type: length # example or length |
|---|
| | | batch_size: 7000 # if batch_type is example, batch_size is the numbers of samples; if length, batch_size is source_token_len+target_token_len; |
|---|
| | | max_token_length: 2000 # filter samples if source_token_len+target_token_len > max_token_length, |
|---|
| | | min_token_length: 60 |
|---|
| | | shuffle: True |
|---|
| | | num_workers: 4 |
|---|
| | | sos: ${model_conf.sos} |
|---|
| | | eos: ${model_conf.eos} |
|---|
| | | |
|---|
| | | train_conf: |
|---|
| | | accum_grad: 2 |
|---|
| | | grad_clip: 5 |
|---|
| | | max_epoch: 20 |
|---|
| | | keep_nbest_models: 20 |
|---|
| | | avg_nbest_model: ${train_conf.keep_nbest_models} |
|---|
| | | log_interval: 50 |
|---|
| | | |
|---|
| | | optim: adamw |
|---|
| | | optim_conf: |
|---|
| | | lr: 0.00002 |
|---|
| | | |
|---|
| | | scheduler: warmuplr |
|---|
| | | scheduler_conf: |
|---|
| | | warmup_steps: 10000 |
|---|
| | | |
|---|
| | | specaug: SpecAug |
|---|
| | | specaug_conf: |
|---|
| | | apply_time_warp: true |
|---|
| | | time_warp_window: 5 |
|---|
| | | time_warp_mode: bicubic |
|---|
| | | apply_freq_mask: true |
|---|
| | | freq_mask_width_range: |
|---|
| | | - 0 |
|---|
| | | - 40 |
|---|
| | | num_freq_mask: 2 |
|---|
| | | apply_time_mask: true |
|---|
| | | time_mask_width_ratio_range: |
|---|
| | | - 0.0 |
|---|
| | | - 0.12 |
|---|
| | | num_time_mask: 2 |
|---|
| | | |
|---|
| | | scope_map: ['encoder.encoders', 'model.encoder', 'decoder.decoders', 'model.decoder'] |
|---|
