Yoshiki's picture
Update model
570a094
metadata
tags:
  - espnet
  - audio
  - speech-enhancement-recognition
language: en
datasets:
  - chime4
license: cc-by-4.0

ESPnet2 EnhS2T model

Yoshiki/chime4_enh_asr1_wpd_wavlm_conformer

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

Demo: How to use in ESPnet2

Follow the ESPnet installation instructions if you haven't done that already.

cd espnet
8ed83f45d5aa2ca6b3635e44b9c29afb9b5fb600
pip install -e .
cd egs2/chime4/enh_asr1
./run.sh --skip_data_prep false --skip_train true --download_model Yoshiki/chime4_enh_asr1_wpd_wavlm_conformer

RESULTS

Environments

  • date: Tue Oct 11 02:40:53 UTC 2022
  • python version: 3.7.4 (default, Aug 13 2019, 20:35:49) [GCC 7.3.0]
  • espnet version: espnet 202207
  • pytorch version: pytorch 1.10.1+cu111
  • Git hash: ``
    • Commit date: ``

enh_asr_train_enh_asr_wpd_init_noenhloss_wavlm_conformer_raw_en_char

WER

dataset Snt Wrd Corr Sub Del Ins Err S.Err
decode_asr_transformer_largelm_normalize_output_wavtrue_lm_lm_train_lm_transformer_en_char_valid.loss.ave_enh_asr_model_valid.acc.ave_10best/dt05_real_isolated_6ch_track 1640 27119 98.8 0.9 0.2 0.2 1.3 16.2
decode_asr_transformer_largelm_normalize_output_wavtrue_lm_lm_train_lm_transformer_en_char_valid.loss.ave_enh_asr_model_valid.acc.ave_10best/dt05_simu_isolated_6ch_track 1640 27120 98.9 0.9 0.2 0.1 1.3 15.2
decode_asr_transformer_largelm_normalize_output_wavtrue_lm_lm_train_lm_transformer_en_char_valid.loss.ave_enh_asr_model_valid.acc.ave_10best/et05_real_isolated_6ch_track 1320 21409 98.4 1.4 0.2 0.2 1.8 20.6
decode_asr_transformer_largelm_normalize_output_wavtrue_lm_lm_train_lm_transformer_en_char_valid.loss.ave_enh_asr_model_valid.acc.ave_10best/et05_simu_isolated_6ch_track 1320 21416 98.9 1.0 0.2 0.1 1.2 15.2

CER

dataset Snt Wrd Corr Sub Del Ins Err S.Err
decode_asr_transformer_largelm_normalize_output_wavtrue_lm_lm_train_lm_transformer_en_char_valid.loss.ave_enh_asr_model_valid.acc.ave_10best/dt05_real_isolated_6ch_track 1640 160390 99.7 0.1 0.2 0.2 0.5 16.2
decode_asr_transformer_largelm_normalize_output_wavtrue_lm_lm_train_lm_transformer_en_char_valid.loss.ave_enh_asr_model_valid.acc.ave_10best/dt05_simu_isolated_6ch_track 1640 160400 99.7 0.1 0.2 0.1 0.5 15.2
decode_asr_transformer_largelm_normalize_output_wavtrue_lm_lm_train_lm_transformer_en_char_valid.loss.ave_enh_asr_model_valid.acc.ave_10best/et05_real_isolated_6ch_track 1320 126796 99.5 0.2 0.3 0.2 0.7 20.6
decode_asr_transformer_largelm_normalize_output_wavtrue_lm_lm_train_lm_transformer_en_char_valid.loss.ave_enh_asr_model_valid.acc.ave_10best/et05_simu_isolated_6ch_track 1320 126812 99.7 0.2 0.2 0.1 0.5 15.2

TER

dataset Snt Wrd Corr Sub Del Ins Err S.Err

EnhS2T config

expand
config: conf/tuning/train_enh_asr_wpd_init_noenhloss_wavlm_conformer.yaml
print_config: false
log_level: INFO
dry_run: false
iterator_type: sequence
output_dir: exp/enh_asr_train_enh_asr_wpd_init_noenhloss_wavlm_conformer_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: 31
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: 1
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
create_graph_in_tensorboard: false
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_beamformer_wpd_ci_sdr_shorttap_raw/valid.loss.best.pth:separator:enh_model.separator
- ../asr1/exp/asr_train_asr_conformer_wavlm2_raw_en_char/valid.acc.best.pth:frontend:s2t_model.frontend
- ../asr1/exp/asr_train_asr_conformer_wavlm2_raw_en_char/valid.acc.best.pth:preencoder:s2t_model.preencoder
- ../asr1/exp/asr_train_asr_conformer_wavlm2_raw_en_char/valid.acc.best.pth:encoder:s2t_model.encoder
- ../asr1/exp/asr_train_asr_conformer_wavlm2_raw_en_char/valid.acc.best.pth:ctc:s2t_model.ctc
- ../asr1/exp/asr_train_asr_conformer_wavlm2_raw_en_char/valid.acc.best.pth:decoder:s2t_model.decoder
ignore_init_mismatch: false
freeze_param:
- s2t_model.frontend.upstream
num_iters_per_epoch: null
batch_size: 16
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_spk1_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_spk1_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_isolated_6ch_track/wav.scp
    - speech
    - sound
-   - dump/raw/tr05_multi_isolated_6ch_track/spk1.scp
    - speech_ref1
    - sound
-   - dump/raw/tr05_multi_isolated_6ch_track/text_spk1
    - text_spk1
    - text
valid_data_path_and_name_and_type:
-   - dump/raw/dt05_multi_isolated_6ch_track/wav.scp
    - speech
    - sound
-   - dump/raw/dt05_multi_isolated_6ch_track/spk1.scp
    - speech_ref1
    - sound
-   - dump/raw/dt05_multi_isolated_6ch_track/text_spk1
    - text_spk1
    - text
allow_variable_data_keys: false
max_cache_size: 0.0
max_cache_fd: 32
valid_max_cache_size: null
optim: sgd
optim_conf:
    lr: 0.001
    momentum: 0.9
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: null
    zero_infinity: true
enh_criterions:
-   name: ci_sdr
    conf:
        filter_length: 512
    wrapper: fixed_order
    wrapper_conf:
        weight: 0.1
diar_num_spk: null
diar_input_size: null
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
diar_model_conf:
    diar_weight: 1.0
    attractor_weight: 1.0
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
text_name:
- text_spk1
enh_encoder: stft
enh_encoder_conf:
    n_fft: 512
    win_length: 400
    hop_length: 128
    use_builtin_complex: false
enh_separator: wpe_beamformer
enh_separator_conf:
    num_spk: 1
    loss_type: spectrum
    use_wpe: false
    wnet_type: blstmp
    wlayers: 3
    wunits: 512
    wprojs: 512
    wdropout_rate: 0.0
    taps: 3
    delay: 3
    use_dnn_mask_for_wpe: true
    use_beamformer: true
    bnet_type: blstmp
    blayers: 3
    bunits: 512
    bprojs: 512
    badim: 320
    ref_channel: 4
    use_noise_mask: true
    beamformer_type: wpd_souden
    bdropout_rate: 0.0
enh_decoder: stft
enh_decoder_conf:
    n_fft: 512
    win_length: 400
    hop_length: 128
enh_mask_module: multi_mask
enh_mask_module_conf: {}
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: 80
asr_encoder: conformer
asr_encoder_conf:
    output_size: 256
    attention_heads: 4
    linear_units: 2048
    num_blocks: 12
    dropout_rate: 0.1
    positional_dropout_rate: 0.1
    attention_dropout_rate: 0.0
    input_layer: conv2d2
    normalize_before: true
    macaron_style: true
    pos_enc_layer_type: rel_pos
    selfattention_layer_type: rel_selfattn
    activation_type: swish
    use_cnn_module: true
    cnn_module_kernel: 15
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: 0.1
    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: {}
diar_frontend: default
diar_frontend_conf: {}
diar_specaug: null
diar_specaug_conf: {}
diar_normalize: utterance_mvn
diar_normalize_conf: {}
diar_encoder: transformer
diar_encoder_conf: {}
diar_decoder: linear
diar_decoder_conf: {}
label_aggregator: label_aggregator
label_aggregator_conf: {}
diar_attractor: null
diar_attractor_conf: {}
required:
- output_dir
version: '202207'
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}
}