ESPnet2 ASR model

`espnet/zuazo_commonvoice_asr_train_asr_conformer5_raw_eu_bpe150_sp`

This model was trained by Xabier de Zuazo using commonvoice 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 5d0758e2a7063b82d1f10a8ac2de98eb6cf8a352
pip install -e .
cd egs2/commonvoice/asr1.conformer.lm.best
./run.sh --skip_data_prep false --skip_train true --download_model espnet/zuazo_commonvoice_asr_train_asr_conformer5_raw_eu_bpe150_sp

RESULTS

Environments

date: Thu Sep 21 09:55:45 CEST 2023
python version: 3.8.17 | packaged by conda-forge | (default, Jun 16 2023, 07:06:00) [GCC 11.4.0]
espnet version: espnet 202308
pytorch version: pytorch 2.0.1
Git hash: 5d0758e2a7063b82d1f10a8ac2de98eb6cf8a352
- Commit date: Wed Aug 30 18:03:42 2023 -0400

exp/asr_train_asr_conformer5_raw_eu_bpe150_sp

WER

dataset	Snt	Wrd	Corr	Sub	Del	Ins	Err	S.Err
decode_asr_lm_lm_train_lm_eu_bpe150_valid.loss.ave_asr_model_valid.acc.ave/test_eu	6640	49267	92.8	6.8	0.4	0.8	8.0	33.3

CER

dataset	Snt	Wrd	Corr	Sub	Del	Ins	Err	S.Err
decode_asr_lm_lm_train_lm_eu_bpe150_valid.loss.ave_asr_model_valid.acc.ave/test_eu	6640	373913	98.8	0.6	0.7	0.4	1.6	33.3

TER

dataset	Snt	Wrd	Corr	Sub	Del	Ins	Err	S.Err
decode_asr_lm_lm_train_lm_eu_bpe150_valid.loss.ave_asr_model_valid.acc.ave/test_eu	6640	208360	97.4	1.5	1.1	0.5	3.1	33.3

exp/asr_train_asr_conformer5_raw_eu_bpe150_sp/decode_asr_lm_lm_train_lm_eu_bpe150_valid.loss.ave_asr_model_valid.acc.ave

WER

dataset	Snt	Wrd	Corr	Sub	Del	Ins	Err	S.Err
org/dev_eu	6640	49505	93.5	6.2	0.3	0.8	7.3	31.0

CER

dataset	Snt	Wrd	Corr	Sub	Del	Ins	Err	S.Err
org/dev_eu	6640	376502	99.0	0.5	0.5	0.3	1.4	31.0

TER

dataset	Snt	Wrd	Corr	Sub	Del	Ins	Err	S.Err
org/dev_eu	6640	209465	97.7	1.4	1.0	0.4	2.7	31.0

ASR config

expand

config: conf/tuning/train_asr_conformer5.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_conformer5_raw_eu_bpe150_sp
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: false
sharded_ddp: false
cudnn_enabled: true
cudnn_benchmark: false
cudnn_deterministic: true
collect_stats: false
write_collected_feats: false
max_epoch: 50
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: 3
grad_clip_type: 2.0
grad_noise: false
accum_grad: 1
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: []
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_eu_bpe150_sp/train/speech_shape
- exp/asr_stats_raw_eu_bpe150_sp/train/text_shape.bpe
valid_shape_file:
- exp/asr_stats_raw_eu_bpe150_sp/valid/speech_shape
- exp/asr_stats_raw_eu_bpe150_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_eu_sp/wav.scp
    - speech
    - sound
-   - dump/raw/train_eu_sp/text
    - text
    - text
valid_data_path_and_name_and_type:
-   - dump/raw/dev_eu/wav.scp
    - speech
    - sound
-   - dump/raw/dev_eu/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: 4.0
scheduler: noamlr
scheduler_conf:
    model_size: 256
    warmup_steps: 25000
token_list:
- <blank>
- <unk>
- A
- ▁
- I
- E
- Z
- .
- R
- N
- U
- S
- O
- T
- KO
- K
- ▁E
- TU
- TE
- RA
- EN
- L
- ','
- LA
- TA
- AK
- ▁A
- AN
- ▁DA
- RE
- KA
- P
- GO
- IN
- B
- M
- ▁DU
- RI
- GU
- ▁ETA
- D
- ER
- UR
- ▁BA
- ▁P
- H
- MA
- ▁G
- ▁I
- ▁HA
- TZEN
- LE
- ▁EZ
- ▁O
- EK
- GI
- ▁BAT
- DA
- DU
- TZA
- KI
- DI
- RO
- ▁GA
- REN
- AR
- TEN
- GA
- TIK
- RRI
- ▁BI
- LI
- ▁BER
- G
- ▁AR
- TO
- ERA
- AREN
- ▁ZI
- ▁DE
- ▁BE
- X
- BA
- ▁DI
- ▁IZAN
- ▁ZE
- ETAN
- ▁ZEN
- EAN
- IA
- ▁JA
- ▁ERE
- ▁DITU
- ▁ZA
- ▁ERA
- LO
- ▁HOR
- NTZ
- ▁DIRA
- MEN
- ▁HI
- ▁F
- F
- LDE
- ZIO
- '?'
- ▁ZU
- '-'
- DO
- ▁EGIN
- TZEKO
- ▁BEHAR
- TZI
- BIL
- ▁IN
- RIK
- ▁HORI
- ▁SA
- ▁NA
- BIDE
- ▁KON
- ▁HE
- ▁ZUEN
- ▁MU
- ▁BESTE
- ▁SO
- ▁HERRI
- ▁IKAS
- ▁NO
- ▁ALD
- ▁NI
- ▁TX
- ABE
- KETA
- ▁BAINA
- C
- '!'
- V
- Y
- ':'
- ;
- '"'
- í
- Q
- ñ
- W
- J
- ‘
- ’
- <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/eu_token_list/bpe_unigram150/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
    win_length: 400
    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
    - 30
    num_freq_mask: 2
    apply_time_mask: true
    time_mask_width_range:
    - 0
    - 40
    num_time_mask: 2
normalize: global_mvn
normalize_conf:
    stats_file: exp/asr_stats_raw_eu_bpe150_sp/train/feats_stats.npz
model: espnet
model_conf:
    ctc_weight: 0.3
    lsm_weight: 0.1
    length_normalized_loss: false
preencoder: null
preencoder_conf: {}
encoder: conformer
encoder_conf:
    input_layer: conv2d
    num_blocks: 12
    linear_units: 2048
    dropout_rate: 0.1
    output_size: 256
    attention_heads: 4
    attention_dropout_rate: 0.0
    pos_enc_layer_type: rel_pos
    selfattention_layer_type: rel_selfattn
    activation_type: swish
    macaron_style: true
    use_cnn_module: true
    cnn_module_kernel: 15
postencoder: null
postencoder_conf: {}
decoder: transformer
decoder_conf:
    input_layer: embed
    num_blocks: 6
    linear_units: 2048
    dropout_rate: 0.1
preprocessor: default
preprocessor_conf: {}
required:
- output_dir
- token_list
version: '202308'
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}
}