ESPnet2 ASR model

`espnet/akreal_libritts_asr_phn`

This model was trained by Pavel Denisov using libritts recipe in espnet.

Demo: How to use in ESPnet2

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

cd espnet
git checkout 3800a13ae8972839d506b85585c41e6b24daf812
pip install -e .
cd egs2/libritts/asr1
./run.sh --skip_data_prep false --skip_train true --download_model espnet/akreal_libritts_asr_phn

RESULTS

Environments

date: Sat Oct 14 03:02:09 CEST 2023
python version: 3.10.8 (main, Nov 14 2022, 00:00:00) [GCC 11.3.1 20220421 (Red Hat 11.3.1-3)]
espnet version: espnet 202308
pytorch version: pytorch 2.0.1+cu118
Git hash: 3800a13ae8972839d506b85585c41e6b24daf812
- Commit date: Sun Oct 8 17:51:17 2023 +0200

exp/asr_train_asr_raw_en_bpe100_sp

WER

dataset	Snt	Wrd	Corr	Sub	Del	Ins	Err	S.Err
decode_asr_asr_model_valid.acc.ave/dev-clean	5736	95872	91.7	8.0	0.4	0.8	9.1	67.0
decode_asr_asr_model_valid.acc.ave/dev-other	4613	69577	88.5	10.9	0.6	1.2	12.7	74.2
decode_asr_asr_model_valid.acc.ave/test-clean	4837	87078	91.4	8.2	0.4	0.8	9.4	70.4
decode_asr_asr_model_valid.acc.ave/test-other	5120	72541	87.0	12.2	0.8	1.1	14.1	77.1

CER

dataset	Snt	Wrd	Corr	Sub	Del	Ins	Err	S.Err
decode_asr_asr_model_valid.acc.ave/dev-clean	5736	570710	98.4	0.8	0.9	0.6	2.2	67.1
decode_asr_asr_model_valid.acc.ave/dev-other	4613	414781	97.2	1.6	1.2	1.0	3.8	74.2
decode_asr_asr_model_valid.acc.ave/test-clean	4837	530647	98.5	0.7	0.8	0.6	2.2	70.5
decode_asr_asr_model_valid.acc.ave/test-other	5120	429463	96.7	1.7	1.6	1.0	4.3	77.1

TER

dataset	Snt	Wrd	Corr	Sub	Del	Ins	Err	S.Err
decode_asr_asr_model_valid.acc.ave/dev-clean	5736	433548	97.6	1.4	1.0	0.6	3.0	67.1
decode_asr_asr_model_valid.acc.ave/dev-other	4613	316550	96.1	2.5	1.4	1.0	5.0	74.2
decode_asr_asr_model_valid.acc.ave/test-clean	4837	404031	97.7	1.4	0.9	0.7	2.9	70.5
decode_asr_asr_model_valid.acc.ave/test-other	5120	327248	95.4	2.8	1.8	1.1	5.7	77.1

ASR config

expand

config: conf/train_asr.yaml
print_config: false
log_level: INFO
drop_last_iter: false
dry_run: false
iterator_type: sequence
valid_iterator_type: null
output_dir: exp/asr_train_asr_raw_en_bpe100_sp
ngpu: 1
seed: 0
num_workers: 1
num_att_plot: 3
dist_backend: nccl
dist_init_method: env://
dist_world_size: 2
dist_rank: 0
local_rank: 0
dist_master_addr: localhost
dist_master_port: 58465
dist_launcher: null
multiprocessing_distributed: true
unused_parameters: true
sharded_ddp: false
cudnn_enabled: true
cudnn_benchmark: false
cudnn_deterministic: true
collect_stats: false
write_collected_feats: false
max_epoch: 70
patience: null
val_scheduler_criterion:
- valid
- loss
early_stopping_criterion:
- valid
- loss
- min
best_model_criterion:
-   - valid
    - acc
    - max
keep_nbest_models: 10
nbest_averaging_interval: 0
grad_clip: 5.0
grad_clip_type: 2.0
grad_noise: false
accum_grad: 8
no_forward_run: false
resume: true
train_dtype: float32
use_amp: true
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: []
ignore_init_mismatch: false
freeze_param: []
num_iters_per_epoch: null
batch_size: 20
valid_batch_size: null
batch_bins: 10000000
valid_batch_bins: null
train_shape_file:
- exp/asr_stats_raw_en_bpe100_sp/train/speech_shape
- exp/asr_stats_raw_en_bpe100_sp/train/text_shape.bpe
valid_shape_file:
- exp/asr_stats_raw_en_bpe100_sp/valid/speech_shape
- exp/asr_stats_raw_en_bpe100_sp/valid/text_shape.bpe
batch_type: numel
valid_batch_type: null
fold_length:
- 80000
- 150
sort_in_batch: descending
shuffle_within_batch: false
sort_batch: descending
multiple_iterator: false
chunk_length: 500
chunk_shift_ratio: 0.5
num_cache_chunks: 1024
chunk_excluded_key_prefixes: []
train_data_path_and_name_and_type:
-   - dump/raw/train-960_sp/wav.scp
    - speech
    - kaldi_ark
-   - dump/raw/train-960_sp/text
    - text
    - text
valid_data_path_and_name_and_type:
-   - dump/raw/dev/wav.scp
    - speech
    - kaldi_ark
-   - dump/raw/dev/text
    - text
    - text
allow_variable_data_keys: false
max_cache_size: 0.0
max_cache_fd: 32
valid_max_cache_size: null
exclude_weight_decay: false
exclude_weight_decay_conf: {}
optim: adam
optim_conf:
    lr: 0.005
    weight_decay: 1.0e-06
scheduler: warmuplr
scheduler_conf:
    warmup_steps: 40000
token_list:
- <blank>
- <unk>
- ˈ
- ː
- ▁
- ɹ
- t
- d
- ɪ
- i
- ˌ
- ɛ
- ','
- s
- l
- n
- k
- z
- m
- ▁s
- eɪ
- ʌ
- aɪ
- .
- ɔ
- æ
- ɚ
- oʊ
- ɑ
- ▁w
- ▁h
- v
- ▁b
- ▁m
- p
- u
- ə
- ▁f
- ▁k
- ▁ɐ
- ▁ðə
- ən
- ▁ð
- ▁ˈ
- ɪŋ
- ▁ænd
- ɜ
- f
- ʊ
- ▁p
- ɾ
- ▁ʌv
- ▁d
- st
- ▁tə
- ɛn
- ▁l
- aʊ
- əl
- b
- ▁n
- ʃ
- ▁t
- tʃ
- ▁ɪn
- ▁ɡ
- ðə
- ɪn
- ▁ɹ
- θ
- w
- '"'
- ▁j
- dʒ
- æn
- ▁"
- ɡ
- ð
- o
- ɐ
- j
- ŋ
- ;
- '?'
- '!'
- h
- ':'
- ʒ
- ʔ
- r
- —
- ɬ
- x
- ç
- ̩
- ᵻ
- e
- a
- ̃
- <sos/eos>
init: null
input_size: null
ctc_conf:
    dropout_rate: 0.0
    ctc_type: builtin
    reduce: true
    ignore_nan_grad: null
    zero_infinity: true
joint_net_conf: null
use_preprocessor: true
token_type: bpe
bpemodel: data/en_token_list/bpe_unigram100/bpe.model
non_linguistic_symbols: null
cleaner: null
g2p: null
speech_volume_normalize: null
rir_scp: null
rir_apply_prob: 1.0
noise_scp: null
noise_apply_prob: 1.0
noise_db_range: '13_15'
short_noise_thres: 0.5
aux_ctc_tasks: []
frontend: default
frontend_conf:
    n_fft: 512
    hop_length: 160
    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
    - 27
    num_freq_mask: 2
    apply_time_mask: true
    time_mask_width_ratio_range:
    - 0.0
    - 0.05
    num_time_mask: 10
normalize: global_mvn
normalize_conf:
    stats_file: exp/asr_stats_raw_en_bpe100_sp/train/feats_stats.npz
model: espnet
model_conf:
    ctc_weight: 0.6
    lsm_weight: 0.1
    length_normalized_loss: false
preencoder: null
preencoder_conf: {}
encoder: e_branchformer
encoder_conf:
    output_size: 512
    attention_heads: 8
    attention_layer_type: rel_selfattn
    pos_enc_layer_type: rel_pos
    rel_pos_type: latest
    cgmlp_linear_units: 3072
    cgmlp_conv_kernel: 31
    use_linear_after_conv: false
    gate_activation: identity
    num_blocks: 17
    dropout_rate: 0.1
    positional_dropout_rate: 0.1
    attention_dropout_rate: 0.1
    input_layer: conv2d
    layer_drop_rate: 0.1
    linear_units: 1024
    positionwise_layer_type: linear
    macaron_ffn: true
    use_ffn: true
    merge_conv_kernel: 31
postencoder: null
postencoder_conf: {}
decoder: transformer
decoder_conf:
    attention_heads: 8
    linear_units: 2048
    num_blocks: 6
    dropout_rate: 0.1
    positional_dropout_rate: 0.1
    self_attention_dropout_rate: 0.1
    src_attention_dropout_rate: 0.1
    layer_drop_rate: 0.2
preprocessor: default
preprocessor_conf: {}
required:
- output_dir
- token_list
version: '202308'
distributed: true

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}
}