ESPnet2 ASR model

`espnet/fsc_challenge_slu_2pass_transformer_gt`

This model was trained by Siddhant using fsc_challenge 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 3b54bfe52a294cdfce668c20d777bfa65f413745
pip install -e .
cd egs2/fsc_challenge/slu1
./run.sh --skip_data_prep false --skip_train true --download_model espnet/fsc_challenge_slu_2pass_transformer_gt

RESULTS

Environments

date: Sun Mar 13 20:59:06 EDT 2022
python version: 3.8.11 (default, Aug 3 2021, 15:09:35) [GCC 7.5.0]
espnet version: espnet 0.10.3a3
pytorch version: pytorch 1.9.0+cu102
Git hash: 97b9dad4dbca71702cb7928a126ec45d96414a3f
- Commit date: Mon Sep 13 22:55:04 2021 +0900

asr_train_asr_hubert_transformer_adam_specaug_deliberation_transformer_3_raw_en_word

WER

dataset	Snt	Wrd	Corr	Sub	Del	Ins	Err	S.Err
inference_asr_model_valid.acc.ave_5best/spk_test	3349	17937	99.9	0.1	0.0	0.0	0.1	0.6
inference_asr_model_valid.acc.ave_5best/utt_test	4204	22540	89.8	6.6	3.6	0.0	10.2	27.6

CER

dataset	Snt	Wrd	Corr	Sub	Del	Ins	Err	S.Err
inference_asr_model_valid.acc.ave_5best/spk_test	3349	152191	100.0	0.0	0.0	0.0	0.1	0.6
inference_asr_model_valid.acc.ave_5best/utt_test	4204	191435	94.5	2.8	2.7	0.5	6.0	27.6

TER

dataset	Snt	Wrd	Corr	Sub	Del	Ins	Err	S.Err

ASR config

expand

config: conf/tuning/train_asr_hubert_transformer_adam_specaug_deliberation_transformer_3.yaml
print_config: false
log_level: INFO
dry_run: false
iterator_type: sequence
output_dir: exp/asr_train_asr_hubert_transformer_adam_specaug_deliberation_transformer_3_raw_en_word
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: 25
patience: null
val_scheduler_criterion:
- valid
- loss
early_stopping_criterion:
- valid
- loss
- min
best_model_criterion:
-   - train
    - loss
    - min
-   - valid
    - loss
    - min
-   - train
    - acc
    - max
-   - valid
    - acc
    - max
keep_nbest_models: 5
grad_clip: 5.0
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_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:
- ../../fsc_challenge/asr1/exp/asr_train_asr_hubert_transformer_adam_specaug_old_raw_en_word/valid.acc.ave_5best.pth:encoder:encoder
ignore_init_mismatch: false
freeze_param:
- encoder
- postdecoder.model
- frontend.upstream
num_iters_per_epoch: null
batch_size: 20
valid_batch_size: null
batch_bins: 1000000
valid_batch_bins: null
train_shape_file:
- exp/asr_stats_raw_en_word/train/speech_shape
- exp/asr_stats_raw_en_word/train/text_shape.word
- exp/asr_stats_raw_en_word/train/transcript_shape.word
valid_shape_file:
- exp/asr_stats_raw_en_word/valid/speech_shape
- exp/asr_stats_raw_en_word/valid/text_shape.word
- exp/asr_stats_raw_en_word/valid/transcript_shape.word
batch_type: folded
valid_batch_type: null
fold_length:
- 80000
- 150
- 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/wav.scp
    - speech
    - sound
-   - dump/raw/train/text
    - text
    - text
-   - dump/raw/train/transcript
    - transcript
    - text
valid_data_path_and_name_and_type:
-   - dump/raw/valid/wav.scp
    - speech
    - sound
-   - dump/raw/valid/text
    - text
    - text
-   - dump/raw/valid/transcript
    - transcript
    - 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: 0.0002
scheduler: warmuplr
scheduler_conf:
    warmup_steps: 25000
token_list:
- <blank>
- <unk>
- the
- turn
- lights
- in
- up
- 'on'
- down
- temperature
- heat
- switch
- kitchen
- volume
- 'off'
- increase_volume_none
- bedroom
- washroom
- decrease_volume_none
- language
- bathroom
- decrease
- my
- to
- increase
- decrease_heat_washroom
- increase_heat_washroom
- music
- heating
- bring
- increase_heat_none
- too
- decrease_heat_none
- me
- change_language_none_none
- activate_lights_washroom
- set
- activate_lights_kitchen
- activate_music_none
- lamp
- deactivate_music_none
- increase_heat_bedroom
- i
- increase_heat_kitchen
- sound
- get
- decrease_heat_kitchen
- loud
- activate_lights_bedroom
- deactivate_lights_bedroom
- decrease_heat_bedroom
- need
- deactivate_lights_kitchen
- bring_newspaper_none
- newspaper
- bring_shoes_none
- shoes
- bring_socks_none
- socks
- activate_lights_none
- deactivate_lights_none
- louder
- go
- deactivate_lights_washroom
- change_language_Chinese_none
- chinese
- could
- you
- bring_juice_none
- juice
- deactivate_lamp_none
- make
- activate_lamp_none
- it
- stop
- play
- change
- quiet
- change_language_Korean_none
- korean
- some
- practice
- change_language_German_none
- german
- ok
- now
- main
- change_language_English_none
- english
- its
- hear
- pause
- this
- thats
- lower
- far
- audio
- please
- fetch
- phones
- a
- different
- start
- resume
- softer
- couldnt
- anything
- quieter
- put
- video
- is
- low
- max
- phone
- mute
- reduce
- use
- languages
- allow
- device
- system
- <sos/eos>
transcript_token_list:
- <blank>
- <unk>
- the
- turn
- lights
- in
- up
- 'on'
- down
- temperature
- heat
- switch
- kitchen
- volume
- 'off'
- bedroom
- washroom
- language
- bathroom
- decrease
- my
- to
- increase
- music
- heating
- bring
- too
- me
- set
- lamp
- i
- sound
- get
- loud
- need
- newspaper
- shoes
- socks
- louder
- go
- chinese
- could
- you
- juice
- make
- it
- stop
- play
- change
- quiet
- korean
- some
- practice
- german
- ok
- now
- main
- english
- its
- hear
- pause
- this
- thats
- lower
- far
- audio
- please
- fetch
- phones
- a
- different
- start
- resume
- softer
- couldnt
- anything
- quieter
- put
- video
- is
- low
- max
- phone
- mute
- reduce
- use
- languages
- allow
- device
- system
- <sos/eos>
two_pass: false
pre_postencoder_norm: false
init: null
input_size: null
ctc_conf:
    dropout_rate: 0.0
    ctc_type: builtin
    reduce: true
    ignore_nan_grad: true
model_conf:
    transcript_token_list:
    - <blank>
    - <unk>
    - the
    - turn
    - lights
    - in
    - up
    - 'on'
    - down
    - temperature
    - heat
    - switch
    - kitchen
    - volume
    - 'off'
    - bedroom
    - washroom
    - language
    - bathroom
    - decrease
    - my
    - to
    - increase
    - music
    - heating
    - bring
    - too
    - me
    - set
    - lamp
    - i
    - sound
    - get
    - loud
    - need
    - newspaper
    - shoes
    - socks
    - louder
    - go
    - chinese
    - could
    - you
    - juice
    - make
    - it
    - stop
    - play
    - change
    - quiet
    - korean
    - some
    - practice
    - german
    - ok
    - now
    - main
    - english
    - its
    - hear
    - pause
    - this
    - thats
    - lower
    - far
    - audio
    - please
    - fetch
    - phones
    - a
    - different
    - start
    - resume
    - softer
    - couldnt
    - anything
    - quieter
    - put
    - video
    - is
    - low
    - max
    - phone
    - mute
    - reduce
    - use
    - languages
    - allow
    - device
    - system
    - <sos/eos>
    ctc_weight: 0.5
    ignore_id: -1
    lsm_weight: 0.0
    length_normalized_loss: false
    report_cer: true
    report_wer: true
    sym_space: <space>
    sym_blank: <blank>
    extract_feats_in_collect_stats: true
    two_pass: false
    pre_postencoder_norm: false
use_preprocessor: true
token_type: word
bpemodel: null
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: s3prl
frontend_conf:
    frontend_conf:
        upstream: hubert_large_ll60k
    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
    - 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: 1024
    output_size: 80
encoder: transformer
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: conv2d
    normalize_before: true
postencoder: null
postencoder_conf: {}
deliberationencoder: transformer
deliberationencoder_conf:
    output_size: 256
    attention_heads: 4
    linear_units: 2048
    num_blocks: 4
    dropout_rate: 0.1
    positional_dropout_rate: 0.1
    attention_dropout_rate: 0.0
    input_layer: linear
    normalize_before: true
decoder: transformer
decoder_conf:
    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
decoder2: rnn
decoder2_conf: {}
postdecoder: hugging_face_transformers
postdecoder_conf:
    model_name_or_path: bert-base-cased
    output_size: 256
required:
- output_dir
- token_list
version: 0.10.3a3
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}
}