ESPnet2 pretrained model, kamo-naoyuki/timit_asr_train_asr_raw_word_valid.acc.ave, fs=16k, lang=noinfo

This model was trained by kamo-naoyuki using timit recipe in espnet.

• Python API

  See https://github.com/espnet/espnet_model_zoo

• Evaluate in the recipe

  git clone https://github.com/espnet/espnet
  cd espnet
  git checkout f157fcd651df34a714ad2bd5d97e46632e010096
  pip install -e .
  cd egs2/timit/asr1
  ./run.sh --skip_data_prep false --skip_train true --download_model kamo-naoyuki/timit_asr_train_asr_raw_word_valid.acc.ave
  

• Results

  
  # RESULTS
  ## Environments
  - date: `Sun Nov 22 05:40:05 JST 2020`
  - python version: `3.8.5 (default, Aug  5 2020, 08:36:46)  [GCC 7.3.0]`
  - espnet version: `espnet 0.9.5`
  - pytorch version: `pytorch 1.6.0`
  - Git hash: `f157fcd651df34a714ad2bd5d97e46632e010096`
    - Commit date: `Sun Nov 22 04:29:17 2020 +0900`
  
  ## asr_train_asr_raw_word
  ### WER
  
  |dataset|Snt|Wrd|Corr|Sub|Del|Ins|Err|S.Err|
  |---|---|---|---|---|---|---|---|---|
  |decode_asr_asr_model_valid.acc.ave/dev|400|15057|84.4|11.2|4.4|2.3|17.9|99.0|
  |decode_asr_asr_model_valid.acc.ave/test|192|7215|83.1|12.4|4.5|2.6|19.5|99.0|
  |decode_asr_lm_lm_train_lm_rnn_word_valid.loss.ave_asr_model_valid.acc.ave/dev|400|15057|83.0|12.2|4.7|2.8|19.8|99.8|
  |decode_asr_lm_lm_train_lm_rnn_word_valid.loss.ave_asr_model_valid.acc.ave/test|192|7215|81.6|13.5|4.9|3.3|21.7|100.0|
  
  ### CER
  
  |dataset|Snt|Wrd|Corr|Sub|Del|Ins|Err|S.Err|
  |---|---|---|---|---|---|---|---|---|
  |decode_asr_asr_model_valid.acc.ave/dev|400|41742|91.0|4.3|4.8|2.8|11.9|99.0|
  |decode_asr_asr_model_valid.acc.ave/test|192|20045|90.4|4.7|4.9|3.2|12.8|99.0|
  |decode_asr_lm_lm_train_lm_rnn_word_valid.loss.ave_asr_model_valid.acc.ave/dev|400|41742|90.2|4.7|5.1|3.3|13.1|99.8|
  |decode_asr_lm_lm_train_lm_rnn_word_valid.loss.ave_asr_model_valid.acc.ave/test|192|20045|89.6|5.3|5.2|3.9|14.4|100.0|

• ASR config

  config: conf/train_asr.yaml
  print_config: false
  log_level: INFO
  dry_run: false
  iterator_type: sequence
  output_dir: exp/asr_train_asr_raw_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
  cudnn_enabled: true
  cudnn_benchmark: false
  cudnn_deterministic: true
  collect_stats: false
  write_collected_feats: false
  max_epoch: 25
  patience: 4
  val_scheduler_criterion:
  - valid
  - loss
  early_stopping_criterion:
  - valid
  - loss
  - min
  best_model_criterion:
  -   - valid
      - acc
      - max
  keep_nbest_models: 3
  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
  unused_parameters: false
  use_tensorboard: true
  use_wandb: false
  pretrain_path: null
  init_param: []
  num_iters_per_epoch: null
  batch_size: 30
  valid_batch_size: null
  batch_bins: 1000000
  valid_batch_bins: null
  train_shape_file:
  - exp/asr_stats_raw_word/train/speech_shape
  - exp/asr_stats_raw_word/train/text_shape.word
  valid_shape_file:
  - exp/asr_stats_raw_word/valid/speech_shape
  - exp/asr_stats_raw_word/valid/text_shape.word
  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/wav.scp
      - speech
      - sound
  -   - dump/raw/train/text
      - text
      - text
  valid_data_path_and_name_and_type:
  -   - dump/raw/dev/wav.scp
      - speech
      - sound
  -   - 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
  optim: adadelta
  optim_conf:
      lr: 1.0
      rho: 0.95
      eps: 1.0e-08
      weight_decay: 0
  scheduler: reducelronplateau
  scheduler_conf:
      mode: min
      factor: 0.5
      patience: 1
  token_list:
  - 
  - 
  - sil
  - ih
  - n
  - s
  - ah
  - l
  - r
  - iy
  - er
  - aa
  - t
  - k
  - z
  - m
  - eh
  - p
  - d
  - dh
  - ae
  - ey
  - w
  - f
  - b
  - v
  - uw
  - ay
  - dx
  - hh
  - ow
  - sh
  - ng
  - g
  - jh
  - y
  - ch
  - th
  - aw
  - uh
  - oy
  - 
  init: xavier_uniform
  input_size: null
  ctc_conf:
      dropout_rate: 0.0
      ctc_type: builtin
      reduce: true
      ignore_nan_grad: false
  model_conf:
      ctc_weight: 0.5
      lsm_weight: 0.1
  use_preprocessor: true
  token_type: word
  bpemodel: null
  non_linguistic_symbols: null
  cleaner: null
  g2p: null
  frontend: default
  frontend_conf:
      fs: 16k
  specaug: null
  specaug_conf: {}
  normalize: global_mvn
  normalize_conf:
      stats_file: exp/asr_stats_raw_word/train/feats_stats.npz
  encoder: rnn
  encoder_conf:
      rnn_type: gru
      bidirectional: true
      use_projection: true
      num_layers: 4
      hidden_size: 320
      output_size: 320
      dropout: 0.2
  decoder: rnn
  decoder_conf:
      rnn_type: lstm
      num_layers: 1
      hidden_size: 320
      dropout: 0.0
      att_conf:
          atype: location
          adim: 320
  required:
  - output_dir
  - token_list
  distributed: false

• LM config

  NONE