ESPnet2 EnhS2T model

`espnet/simpleoier_chime4_enh_asr_convtasnet_init_noenhloss_wavlm_transformer_init_raw_en_char`

This model was trained by simpleoier using chime4 recipe in espnet.

Demo: How to use in ESPnet2

cd espnet
git checkout 2b663318cd1773fb8685b1e03295b6bc6889c283
pip install -e .
cd egs2/chime4/enh_asr1
./run.sh --skip_data_prep false --skip_train true --download_model espnet/simpleoier_chime4_enh_asr_convtasnet_init_noenhloss_wavlm_transformer_init_raw_en_char

RESULTS

Environments

date: Thu Apr 28 08:15:30 EDT 2022
python version: 3.7.11 (default, Jul 27 2021, 14:32:16) [GCC 7.5.0]
espnet version: espnet 202204
pytorch version: pytorch 1.8.1
Git hash: ``
- Commit date: ``

enh_asr_train_enh_asr_convtasnet_init_noenhloss_wavlm_transformer_init_lr1e-4_accum1_adam_specaug_bypass0_raw_en_char

WER

dataset	Snt	Wrd	Corr	Sub	Del	Ins	Err	S.Err
decode_asr_transformer_normalize_output_wavtrue_lm_lm_train_lm_transformer_en_char_valid.loss.ave_enh_asr_model_valid.acc.ave/dt05_real_beamformit_2mics	1640	27119	98.5	1.2	0.3	0.2	1.7	19.6
decode_asr_transformer_normalize_output_wavtrue_lm_lm_train_lm_transformer_en_char_valid.loss.ave_enh_asr_model_valid.acc.ave/dt05_real_beamformit_5mics	1640	27119	98.6	1.1	0.3	0.2	1.5	18.7
decode_asr_transformer_normalize_output_wavtrue_lm_lm_train_lm_transformer_en_char_valid.loss.ave_enh_asr_model_valid.acc.ave/dt05_real_isolated_1ch_track	1640	27119	98.3	1.3	0.4	0.2	1.9	21.8
decode_asr_transformer_normalize_output_wavtrue_lm_lm_train_lm_transformer_en_char_valid.loss.ave_enh_asr_model_valid.acc.ave/dt05_simu_beamformit_2mics	1640	27120	97.9	1.5	0.5	0.2	2.3	25.2
decode_asr_transformer_normalize_output_wavtrue_lm_lm_train_lm_transformer_en_char_valid.loss.ave_enh_asr_model_valid.acc.ave/dt05_simu_beamformit_5mics	1640	27120	98.4	1.2	0.4	0.1	1.7	19.9
decode_asr_transformer_normalize_output_wavtrue_lm_lm_train_lm_transformer_en_char_valid.loss.ave_enh_asr_model_valid.acc.ave/dt05_simu_isolated_1ch_track	1640	27120	97.2	2.1	0.7	0.3	3.1	28.9
decode_asr_transformer_normalize_output_wavtrue_lm_lm_train_lm_transformer_en_char_valid.loss.ave_enh_asr_model_valid.acc.ave/et05_real_beamformit_2mics	1320	21409	97.4	2.0	0.6	0.3	2.9	27.3
decode_asr_transformer_normalize_output_wavtrue_lm_lm_train_lm_transformer_en_char_valid.loss.ave_enh_asr_model_valid.acc.ave/et05_real_beamformit_5mics	1320	21409	97.8	1.8	0.4	0.2	2.5	24.3
decode_asr_transformer_normalize_output_wavtrue_lm_lm_train_lm_transformer_en_char_valid.loss.ave_enh_asr_model_valid.acc.ave/et05_real_isolated_1ch_track	1320	21409	96.7	2.6	0.7	0.4	3.7	31.6
decode_asr_transformer_normalize_output_wavtrue_lm_lm_train_lm_transformer_en_char_valid.loss.ave_enh_asr_model_valid.acc.ave/et05_simu_beamformit_2mics	1320	21416	96.6	2.5	1.0	0.3	3.7	32.5
decode_asr_transformer_normalize_output_wavtrue_lm_lm_train_lm_transformer_en_char_valid.loss.ave_enh_asr_model_valid.acc.ave/et05_simu_beamformit_5mics	1320	21416	97.5	1.9	0.7	0.3	2.9	28.9
decode_asr_transformer_normalize_output_wavtrue_lm_lm_train_lm_transformer_en_char_valid.loss.ave_enh_asr_model_valid.acc.ave/et05_simu_isolated_1ch_track	1320	21416	94.6	3.7	1.6	0.5	5.9	37.3

CER

dataset	Snt	Wrd	Corr	Sub	Del	Ins	Err	S.Err
decode_asr_transformer_normalize_output_wavtrue_lm_lm_train_lm_transformer_en_char_valid.loss.ave_enh_asr_model_valid.acc.ave/dt05_real_beamformit_2mics	1640	160390	99.5	0.2	0.3	0.2	0.7	19.6
decode_asr_transformer_normalize_output_wavtrue_lm_lm_train_lm_transformer_en_char_valid.loss.ave_enh_asr_model_valid.acc.ave/dt05_real_beamformit_5mics	1640	160390	99.6	0.1	0.3	0.2	0.6	18.7
decode_asr_transformer_normalize_output_wavtrue_lm_lm_train_lm_transformer_en_char_valid.loss.ave_enh_asr_model_valid.acc.ave/dt05_real_isolated_1ch_track	1640	160390	99.4	0.2	0.4	0.2	0.8	21.8
decode_asr_transformer_normalize_output_wavtrue_lm_lm_train_lm_transformer_en_char_valid.loss.ave_enh_asr_model_valid.acc.ave/dt05_simu_beamformit_2mics	1640	160400	99.2	0.3	0.5	0.2	1.1	25.2
decode_asr_transformer_normalize_output_wavtrue_lm_lm_train_lm_transformer_en_char_valid.loss.ave_enh_asr_model_valid.acc.ave/dt05_simu_beamformit_5mics	1640	160400	99.5	0.2	0.3	0.1	0.7	19.9
decode_asr_transformer_normalize_output_wavtrue_lm_lm_train_lm_transformer_en_char_valid.loss.ave_enh_asr_model_valid.acc.ave/dt05_simu_isolated_1ch_track	1640	160400	98.8	0.5	0.7	0.3	1.5	28.9
decode_asr_transformer_normalize_output_wavtrue_lm_lm_train_lm_transformer_en_char_valid.loss.ave_enh_asr_model_valid.acc.ave/et05_real_beamformit_2mics	1320	126796	98.9	0.4	0.7	0.3	1.4	27.3
decode_asr_transformer_normalize_output_wavtrue_lm_lm_train_lm_transformer_en_char_valid.loss.ave_enh_asr_model_valid.acc.ave/et05_real_beamformit_5mics	1320	126796	99.1	0.4	0.5	0.2	1.1	24.3
decode_asr_transformer_normalize_output_wavtrue_lm_lm_train_lm_transformer_en_char_valid.loss.ave_enh_asr_model_valid.acc.ave/et05_real_isolated_1ch_track	1320	126796	98.6	0.6	0.8	0.4	1.8	31.7
decode_asr_transformer_normalize_output_wavtrue_lm_lm_train_lm_transformer_en_char_valid.loss.ave_enh_asr_model_valid.acc.ave/et05_simu_beamformit_2mics	1320	126812	98.2	0.6	1.1	0.4	2.1	32.5
decode_asr_transformer_normalize_output_wavtrue_lm_lm_train_lm_transformer_en_char_valid.loss.ave_enh_asr_model_valid.acc.ave/et05_simu_beamformit_5mics	1320	126812	98.8	0.4	0.8	0.3	1.5	28.9
decode_asr_transformer_normalize_output_wavtrue_lm_lm_train_lm_transformer_en_char_valid.loss.ave_enh_asr_model_valid.acc.ave/et05_simu_isolated_1ch_track	1320	126812	97.0	1.2	1.9	0.6	3.7	37.3

EnhS2T config

expand

config: conf/tuning/train_enh_asr_convtasnet_init_noenhloss_wavlm_transformer_init_lr1e-4_accum1_adam_specaug_bypass0.yaml
print_config: false
log_level: INFO
dry_run: false
iterator_type: sequence
output_dir: exp/enh_asr_train_enh_asr_convtasnet_init_noenhloss_wavlm_transformer_init_lr1e-4_accum1_adam_specaug_bypass0_raw_en_char
ngpu: 1
seed: 0
num_workers: 1
num_att_plot: 3
dist_backend: nccl
dist_init_method: env://
dist_world_size: null
dist_rank: null
local_rank: 0
dist_master_addr: null
dist_master_port: null
dist_launcher: null
multiprocessing_distributed: false
unused_parameters: true
sharded_ddp: false
cudnn_enabled: true
cudnn_benchmark: false
cudnn_deterministic: true
collect_stats: false
write_collected_feats: false
max_epoch: 12
patience: 10
val_scheduler_criterion:
- valid
- loss
early_stopping_criterion:
- valid
- loss
- min
best_model_criterion:
-   - valid
    - acc
    - max
-   - train
    - loss
    - min
keep_nbest_models: 10
nbest_averaging_interval: 0
grad_clip: 5
grad_clip_type: 2.0
grad_noise: false
accum_grad: 2
no_forward_run: false
resume: true
train_dtype: float32
use_amp: false
log_interval: null
use_matplotlib: true
use_tensorboard: true
use_wandb: false
wandb_project: null
wandb_id: null
wandb_entity: null
wandb_name: null
wandb_model_log_interval: -1
detect_anomaly: false
pretrain_path: null
init_param:
- ../enh1/exp/enh_train_enh_convtasnet_small_raw/valid.loss.ave_1best.pth:encoder:enh_model.encoder
- ../enh1/exp/enh_train_enh_convtasnet_small_raw/valid.loss.ave_1best.pth:separator:enh_model.separator
- ../enh1/exp/enh_train_enh_convtasnet_small_raw/valid.loss.ave_1best.pth:decoder:enh_model.decoder
- ../asr1/exp/asr_train_asr_transformer_wavlm_lr1e-3_specaug_accum1_preenc128_warmup20k_raw_en_char/valid.acc.ave.pth:frontend:s2t_model.frontend
- ../asr1/exp/asr_train_asr_transformer_wavlm_lr1e-3_specaug_accum1_preenc128_warmup20k_raw_en_char/valid.acc.ave.pth:preencoder:s2t_model.preencoder
- ../asr1/exp/asr_train_asr_transformer_wavlm_lr1e-3_specaug_accum1_preenc128_warmup20k_raw_en_char/valid.acc.ave.pth:encoder:s2t_model.encoder
- ../asr1/exp/asr_train_asr_transformer_wavlm_lr1e-3_specaug_accum1_preenc128_warmup20k_raw_en_char/valid.acc.ave.pth:ctc:s2t_model.ctc
- ../asr1/exp/asr_train_asr_transformer_wavlm_lr1e-3_specaug_accum1_preenc128_warmup20k_raw_en_char/valid.acc.ave.pth:decoder:s2t_model.decoder
ignore_init_mismatch: false
freeze_param:
- s2t_model.frontend.upstream
num_iters_per_epoch: null
batch_size: 12
valid_batch_size: null
batch_bins: 1000000
valid_batch_bins: null
train_shape_file:
- exp/enh_asr_stats_raw_en_char/train/speech_shape
- exp/enh_asr_stats_raw_en_char/train/speech_ref1_shape
- exp/enh_asr_stats_raw_en_char/train/text_shape.char
valid_shape_file:
- exp/enh_asr_stats_raw_en_char/valid/speech_shape
- exp/enh_asr_stats_raw_en_char/valid/speech_ref1_shape
- exp/enh_asr_stats_raw_en_char/valid/text_shape.char
batch_type: folded
valid_batch_type: null
fold_length:
- 80000
- 80000
- 150
sort_in_batch: descending
sort_batch: descending
multiple_iterator: false
chunk_length: 500
chunk_shift_ratio: 0.5
num_cache_chunks: 1024
train_data_path_and_name_and_type:
-   - dump/raw/tr05_multi_noisy_si284/wav.scp
    - speech
    - sound
-   - dump/raw/tr05_multi_noisy_si284/spk1.scp
    - speech_ref1
    - sound
-   - dump/raw/tr05_multi_noisy_si284/text
    - text
    - text
valid_data_path_and_name_and_type:
-   - dump/raw/dt05_multi_isolated_1ch_track/wav.scp
    - speech
    - sound
-   - dump/raw/dt05_multi_isolated_1ch_track/spk1.scp
    - speech_ref1
    - sound
-   - dump/raw/dt05_multi_isolated_1ch_track/text
    - text
    - text
allow_variable_data_keys: false
max_cache_size: 0.0
max_cache_fd: 32
valid_max_cache_size: null
optim: adam
optim_conf:
    lr: 0.0001
scheduler: null
scheduler_conf: {}
token_list: data/en_token_list/char/tokens.txt
src_token_list: null
init: xavier_uniform
input_size: null
ctc_conf:
    dropout_rate: 0.0
    ctc_type: builtin
    reduce: true
    ignore_nan_grad: true
enh_criterions:
-   name: si_snr
    conf: {}
    wrapper: fixed_order
    wrapper_conf: {}
enh_model_conf:
    stft_consistency: false
    loss_type: mask_mse
    mask_type: null
asr_model_conf:
    ctc_weight: 0.3
    lsm_weight: 0.1
    length_normalized_loss: false
    extract_feats_in_collect_stats: false
st_model_conf:
    stft_consistency: false
    loss_type: mask_mse
    mask_type: null
subtask_series:
- enh
- asr
model_conf:
    calc_enh_loss: false
    bypass_enh_prob: 0.0
use_preprocessor: true
token_type: char
bpemodel: null
src_token_type: bpe
src_bpemodel: null
non_linguistic_symbols: data/nlsyms.txt
cleaner: null
g2p: null
enh_encoder: conv
enh_encoder_conf:
    channel: 256
    kernel_size: 40
    stride: 20
enh_separator: tcn
enh_separator_conf:
    num_spk: 1
    layer: 4
    stack: 2
    bottleneck_dim: 256
    hidden_dim: 512
    kernel: 3
    causal: false
    norm_type: gLN
    nonlinear: relu
enh_decoder: conv
enh_decoder_conf:
    channel: 256
    kernel_size: 40
    stride: 20
frontend: s3prl
frontend_conf:
    frontend_conf:
        upstream: wavlm_large
    download_dir: ./hub
    multilayer_feature: true
    fs: 16k
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
    - 100
    num_freq_mask: 4
    apply_time_mask: true
    time_mask_width_range:
    - 0
    - 40
    num_time_mask: 2
normalize: utterance_mvn
normalize_conf: {}
asr_preencoder: linear
asr_preencoder_conf:
    input_size: 1024
    output_size: 128
asr_encoder: transformer
asr_encoder_conf:
    output_size: 256
    attention_heads: 4
    linear_units: 2048
    num_blocks: 12
    dropout_rate: 0.1
    attention_dropout_rate: 0.0
    input_layer: conv2d2
    normalize_before: true
asr_postencoder: null
asr_postencoder_conf: {}
asr_decoder: transformer
asr_decoder_conf:
    input_layer: embed
    attention_heads: 4
    linear_units: 2048
    num_blocks: 6
    dropout_rate: 0.1
    positional_dropout_rate: 0.0
    self_attention_dropout_rate: 0.0
    src_attention_dropout_rate: 0.0
st_preencoder: null
st_preencoder_conf: {}
st_encoder: rnn
st_encoder_conf: {}
st_postencoder: null
st_postencoder_conf: {}
st_decoder: rnn
st_decoder_conf: {}
st_extra_asr_decoder: rnn
st_extra_asr_decoder_conf: {}
st_extra_mt_decoder: rnn
st_extra_mt_decoder_conf: {}
required:
- output_dir
- token_list
version: '202204'
distributed: false

Citing ESPnet

@inproceedings{watanabe2018espnet,
  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},
  title={{ESPnet}: End-to-End Speech Processing Toolkit},
  year={2018},
  booktitle={Proceedings of Interspeech},
  pages={2207--2211},
  doi={10.21437/Interspeech.2018-1456},
  url={http://dx.doi.org/10.21437/Interspeech.2018-1456}
}

or arXiv:

@misc{watanabe2018espnet,
  title={ESPnet: End-to-End Speech Processing Toolkit}, 
  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},
  year={2018},
  eprint={1804.00015},
  archivePrefix={arXiv},
  primaryClass={cs.CL}
}