ESPnet2 model

``

This model was trained by Chaitanya Narisetty using recipe in espnet.

Demo: How to use in ESPnet2

cd espnet

pip install -e .
cd egs2/ms_indic_is18/asr1
./run.sh --skip_data_prep false --skip_train true --download_model espnet/chai_microsoft_indian_langs_te

RESULTS

Environments

date: Tue Mar 22 13:38:24 EDT 2022
python version: 3.9.5 (default, Jun 4 2021, 12:28:51) [GCC 7.5.0]
espnet version: espnet 0.10.7a1
pytorch version: pytorch 1.8.1+cu111
Git hash: f91410f712d1287cd6809c5bf26b54c5a40fe314
- Commit date: Mon Mar 14 22:32:17 2022 -0400

asr_train_asr_xlsr53_conformer_raw_te_bpe150_sp

WER

dataset	Snt	Wrd	Corr	Sub	Del	Ins	Err	S.Err
decode_transformer5_lm_lm_train_lm_rnn_te_bpe150_valid.loss.ave_asr_model_valid.acc.ave/test_te	3040	28413	78.0	19.5	2.5	2.4	24.4	80.1
decode_transformer5_lm_lm_train_lm_rnn_te_bpe150_valid.loss.best_asr_model_valid.acc.ave/test_te	3040	28413	78.0	19.4	2.6	2.4	24.4	79.7
decode_transformer5_lm_lm_train_lm_transformer_te_bpe150_valid.loss.ave_asr_model_valid.acc.ave/test_te	3040	28413	78.0	19.5	2.6	2.5	24.5	79.9

CER

dataset	Snt	Wrd	Corr	Sub	Del	Ins	Err	S.Err
decode_transformer5_lm_lm_train_lm_rnn_te_bpe150_valid.loss.ave_asr_model_valid.acc.ave/test_te	3040	229419	95.6	2.2	2.2	1.6	6.1	80.1
decode_transformer5_lm_lm_train_lm_rnn_te_bpe150_valid.loss.best_asr_model_valid.acc.ave/test_te	3040	229419	95.6	2.2	2.2	1.6	6.0	79.7
decode_transformer5_lm_lm_train_lm_transformer_te_bpe150_valid.loss.ave_asr_model_valid.acc.ave/test_te	3040	229419	95.6	2.1	2.2	1.6	6.0	79.9

TER

dataset	Snt	Wrd	Corr	Sub	Del	Ins	Err	S.Err
decode_transformer5_lm_lm_train_lm_rnn_te_bpe150_valid.loss.ave_asr_model_valid.acc.ave/test_te	3040	146657	92.7	4.7	2.6	1.6	8.9	80.1
decode_transformer5_lm_lm_train_lm_rnn_te_bpe150_valid.loss.best_asr_model_valid.acc.ave/test_te	3040	146657	92.8	4.7	2.6	1.6	8.9	79.7
decode_transformer5_lm_lm_train_lm_transformer_te_bpe150_valid.loss.ave_asr_model_valid.acc.ave/test_te	3040	146657	92.8	4.6	2.6	1.6	8.9	79.9

config

expand

config: conf/tuning/train_asr_xlsr53_conformer.yaml
print_config: false
log_level: INFO
dry_run: false
iterator_type: sequence
output_dir: exp/asr_train_asr_xlsr53_conformer_raw_te_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: 15
val_scheduler_criterion:
- valid
- loss
early_stopping_criterion:
- valid
- loss
- min
best_model_criterion:
-   - valid
    - acc
    - max
keep_nbest_models: 5
nbest_averaging_interval: 0
grad_clip: 5
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
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:
- frontend.upstream
num_iters_per_epoch: null
batch_size: 64
valid_batch_size: null
batch_bins: 1000000
valid_batch_bins: null
train_shape_file:
- exp/asr_stats_raw_te_bpe150_sp_ssl/train/speech_shape
- exp/asr_stats_raw_te_bpe150_sp_ssl/train/text_shape.bpe
valid_shape_file:
- exp/asr_stats_raw_te_bpe150_sp_ssl/valid/speech_shape
- exp/asr_stats_raw_te_bpe150_sp_ssl/valid/text_shape.bpe
batch_type: folded
valid_batch_type: null
fold_length:
- 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/train_te_sp/wav.scp
    - speech
    - sound
-   - dump/raw/train_te_sp/text
    - text
    - text
valid_data_path_and_name_and_type:
-   - dump/raw/dev_te/wav.scp
    - speech
    - sound
-   - dump/raw/dev_te/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: 0.0005
scheduler: warmuplr
scheduler_conf:
    warmup_steps: 30000
token_list:
- <blank>
- <unk>
- ా
- ు
- ి
- ం
- ే
- వ
- న
- ల
- ▁అ
- క
- ్
- ో
- మ
- ▁
- త
- ర
- ప
- ీ
- ▁మ
- య
- డ
- ▁ప
- ద
- ని
- గ
- ▁వ
- స
- కు
- ె
- ర్
- ▁స
- ▁క
- ్య
- న్న
- ట
- ▁చ
- ▁త
- ాల
- ంట
- ూ
- శ
- ంద
- ార
- ▁న
- ారు
- ▁ఉ
- లు
- ▁ఆ
- ను
- జ
- రి
- ▁ప్ర
- ించ
- ధ
- ై
- హ
- ంది
- ్ర
- ▁ఇ
- చ
- రు
- స్త
- లో
- ▁ద
- డు
- ▁ఎ
- ▁వి
- ల్ల
- ణ
- గా
- ది
- డి
- న్నారు
- దు
- ిన
- ▁ర
- త్
- ొ
- ▁గ
- ంత
- ంగా
- ▁కా
- బ
- ▁జ
- ష
- ▁తెల
- ులు
- ▁ఏ
- ట్ట
- చ్చ
- తి
- నే
- కి
- ంలో
- ▁అవును
- ▁చెప్ప
- భ
- ▁ఈ
- ప్ప
- ▁ని
- ▁రా
- క్క
- ▁బ
- ట్ల
- ▁భ
- తో
- ▁కూడా
- ▁బా
- ద్ద
- ▁చేస
- ▁లే
- ాయి
- ానికి
- త్ర
- ▁కొ
- ఖ
- ▁ఒక
- ▁చాలా
- క్ష
- ళ
- ▁చేస్త
- ృ
- థ
- ఘ
- ఫ
- ఓ
- ౌ
- ఒ
- ఐ
- ఠ
- ఢ
- అ
- ఉ
- ఏ
- ఈ
- ౦
- ఇ
- ః
- ఋ
- ఝ
- ఔ
- ఛ
- ఞ
- ఊ
- ఎ
- ఆ
- ఙ
- <sos/eos>
init: xavier_uniform
input_size: null
ctc_conf:
    dropout_rate: 0.0
    ctc_type: builtin
    reduce: true
    ignore_nan_grad: true
joint_net_conf: null
model_conf:
    ctc_weight: 0.3
    lsm_weight: 0.1
    length_normalized_loss: false
    extract_feats_in_collect_stats: false
use_preprocessor: true
token_type: bpe
bpemodel: data/te_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'
frontend: fused
frontend_conf:
    frontends:
    -   frontend_type: default
        n_fft: 512
        win_length: 400
        hop_length: 160
    -   frontend_type: s3prl
        frontend_conf:
            upstream: wav2vec2_xlsr
        download_dir: ./hub
        multilayer_feature: true
    align_method: linear_projection
    proj_dim: 200
    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: utterance_mvn
normalize_conf: {}
preencoder: linear
preencoder_conf:
    input_size: 400
    output_size: 100
encoder: conformer
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: 0.1
    input_layer: conv2d
    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
postencoder: null
postencoder_conf: {}
decoder: transformer
decoder_conf:
    input_layer: embed
    num_blocks: 6
    linear_units: 2048
    dropout_rate: 0.1
    positional_dropout_rate: 0.1
    self_attention_dropout_rate: 0.1
    src_attention_dropout_rate: 0.1
required:
- output_dir
- token_list
version: 0.10.7a1
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}
}