Published January 4, 2021 | Version v1
Other Open

ESPnet2 pretrained model, kamo-naoyuki/dirha_wsj_asr_train_asr_transformer_cmvn_raw_char_rir_scpdatadirha_irwav.scp_noise_db_range10_17_noise_scpdatadirha_noisewav.scp_speech_volume_normalize1.0_num_workers2_rir_apply_prob1._sp_valid.acc.ave, fs=16k, lang=en

Creators

Description

This model was trained by kamo-naoyuki using dirha_wsj 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 c30ce885aef4c0eabaa4c5956108c65038b716eb
    pip install -e .
    cd egs2/dirha_wsj/asr1
    ./run.sh --skip_data_prep false --skip_train true --download_model kamo-naoyuki/dirha_wsj_asr_train_asr_transformer_cmvn_raw_char_rir_scpdatadirha_irwav.scp_noise_db_range10_17_noise_scpdatadirha_noisewav.scp_speech_volume_normalize1.0_num_workers2_rir_apply_prob1._sp_valid.acc.ave
    
  • Results
    
    # RESULTS
    ## Environments
    - date: `Sun Dec 27 01:58:16 JST 2020`
    - python version: `3.7.3 (default, Mar 27 2019, 22:11:17)  [GCC 7.3.0]`
    - espnet version: `espnet 0.9.5`
    - pytorch version: `pytorch 1.5.1`
    - Git hash: `bb6bfd7e058998a34365d59c8c48ad57716e8022`
      - Commit date: `Mon Dec 7 11:05:14 2020 +0900`
    
    ## asr_train_asr_transformer_cmvn_raw_char_rir_scpdatadirha_irwav.scp_noise_db_range10_17_noise_scpdatadirha_noisewav.scp_speech_volume_normalize1.0_num_workers2_rir_apply_prob1._sp
    ### WER
    
    |dataset|Snt|Wrd|Corr|Sub|Del|Ins|Err|S.Err|
    |---|---|---|---|---|---|---|---|---|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_real_Kitchen_Circular_Array_KA1|409|6762|72.7|22.2|5.1|3.7|31.0|76.3|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_real_Kitchen_Circular_Array_KA2|409|6762|73.0|22.0|5.0|3.8|30.8|77.8|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_real_Kitchen_Circular_Array_KA3|409|6762|73.5|21.6|4.9|4.0|30.5|76.3|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_real_Kitchen_Circular_Array_KA4|409|6762|72.7|22.2|5.1|3.6|30.9|76.8|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_real_Kitchen_Circular_Array_KA5|409|6762|72.8|22.2|5.0|4.0|31.2|75.3|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_real_Kitchen_Circular_Array_KA6|409|6762|72.3|22.6|5.2|3.4|31.1|76.5|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_real_Livingroom_Circular_Array_Beam_Circular_Array|409|6762|89.5|9.1|1.3|1.4|11.8|49.9|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_real_Livingroom_Circular_Array_LA1|409|6762|88.5|10.2|1.3|1.8|13.4|54.5|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_real_Livingroom_Circular_Array_LA2|409|6762|87.9|10.6|1.4|1.8|13.9|56.7|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_real_Livingroom_Circular_Array_LA3|409|6762|87.8|10.9|1.3|1.9|14.1|55.7|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_real_Livingroom_Circular_Array_LA4|409|6762|88.0|10.5|1.5|2.1|14.1|56.7|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_real_Livingroom_Circular_Array_LA5|409|6762|88.3|10.5|1.2|1.8|13.5|56.0|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_real_Livingroom_Circular_Array_LA6|409|6762|87.7|10.6|1.6|2.1|14.3|55.3|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_real_Livingroom_Linear_Array_Beam_Linear_Array|409|6762|91.7|7.2|1.1|0.8|9.1|50.6|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_real_Livingroom_Linear_Array_LD02|409|6762|87.9|10.3|1.7|2.1|14.1|59.2|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_real_Livingroom_Linear_Array_LD03|409|6762|88.8|9.7|1.5|1.7|12.9|56.0|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_real_Livingroom_Linear_Array_LD04|409|6762|90.4|8.3|1.3|1.0|10.7|53.8|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_real_Livingroom_Linear_Array_LD05|409|6762|87.8|10.4|1.7|1.9|14.0|59.9|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_real_Livingroom_Linear_Array_LD06|409|6762|88.8|9.7|1.5|1.7|12.9|58.7|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_real_Livingroom_Linear_Array_LD07|409|6762|88.6|9.8|1.5|1.6|13.0|58.2|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_real_Livingroom_Linear_Array_LD08|409|6762|86.5|11.7|1.8|2.5|16.0|58.2|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_real_Livingroom_Linear_Array_LD09|409|6762|90.2|8.2|1.6|1.4|11.2|57.7|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_real_Livingroom_Linear_Array_LD10|409|6762|89.0|9.3|1.6|1.6|12.5|58.7|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_real_Livingroom_Linear_Array_LD11|409|6762|89.5|9.1|1.4|1.5|12.1|55.0|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_real_Livingroom_Linear_Array_LD12|409|6762|88.8|9.7|1.4|1.4|12.6|56.0|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_real_Livingroom_Wall_L1C|409|6762|88.4|10.2|1.4|1.6|13.2|54.8|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_real_Livingroom_Wall_L1L|409|6762|88.7|9.9|1.4|1.5|12.8|53.5|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_real_Livingroom_Wall_L1R|409|6762|88.2|10.2|1.5|1.6|13.3|57.5|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_real_Livingroom_Wall_L2L|409|6762|88.1|10.4|1.5|1.7|13.6|53.8|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_real_Livingroom_Wall_L2R|409|6762|87.2|11.3|1.5|2.1|14.9|57.2|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_real_Livingroom_Wall_L3L|409|6762|88.5|10.1|1.3|1.5|13.0|54.3|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_real_Livingroom_Wall_L3R|409|6762|88.6|10.0|1.4|1.4|12.8|55.3|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_real_Livingroom_Wall_L4L|409|6762|88.8|9.6|1.6|1.5|12.7|55.3|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_real_Livingroom_Wall_L4R|409|6762|88.8|9.8|1.4|1.6|12.8|55.0|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_sim_Kitchen_Circular_Array_KA1|409|6766|81.7|16.1|2.2|3.1|21.5|69.4|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_sim_Kitchen_Circular_Array_KA2|409|6766|81.6|16.1|2.4|3.0|21.4|67.7|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_sim_Kitchen_Circular_Array_KA3|409|6766|81.9|15.7|2.3|3.2|21.2|68.2|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_sim_Kitchen_Circular_Array_KA4|409|6766|81.4|16.4|2.2|3.6|22.3|67.7|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_sim_Kitchen_Circular_Array_KA5|409|6766|80.8|16.9|2.3|3.5|22.7|68.2|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_sim_Kitchen_Circular_Array_KA6|409|6766|81.9|16.1|2.0|3.4|21.5|68.0|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_sim_Livingroom_Circular_Array_Beam_Circular_Array|409|6766|92.7|6.5|0.8|0.7|8.0|46.2|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_sim_Livingroom_Circular_Array_LA1|409|6766|91.8|7.4|0.9|0.9|9.1|48.9|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_sim_Livingroom_Circular_Array_LA2|409|6766|91.7|7.4|1.0|0.8|9.1|50.6|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_sim_Livingroom_Circular_Array_LA3|409|6766|91.9|7.2|0.9|0.7|8.8|49.1|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_sim_Livingroom_Circular_Array_LA4|409|6766|91.2|7.7|1.1|0.9|9.7|52.6|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_sim_Livingroom_Circular_Array_LA5|409|6766|91.8|7.3|0.9|0.9|9.1|49.9|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_sim_Livingroom_Circular_Array_LA6|409|6766|91.7|7.3|1.0|1.0|9.3|49.9|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_sim_Livingroom_Linear_Array_Beam_Linear_Array|409|6766|92.9|6.2|0.8|0.5|7.6|47.4|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_sim_Livingroom_Linear_Array_LD02|409|6766|92.0|6.9|1.1|0.6|8.6|49.1|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_sim_Livingroom_Linear_Array_LD03|409|6766|92.5|6.6|0.9|0.8|8.4|48.2|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_sim_Livingroom_Linear_Array_LD04|409|6766|91.9|7.3|0.8|0.8|8.8|50.4|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_sim_Livingroom_Linear_Array_LD05|409|6766|92.3|6.7|1.1|0.7|8.5|48.9|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_sim_Livingroom_Linear_Array_LD06|409|6766|91.9|7.0|1.0|0.8|8.9|48.9|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_sim_Livingroom_Linear_Array_LD07|409|6766|92.3|6.7|1.0|0.7|8.4|49.1|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_sim_Livingroom_Linear_Array_LD08|409|6766|92.4|6.8|0.9|0.7|8.3|47.9|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_sim_Livingroom_Linear_Array_LD09|409|6766|91.8|7.2|1.0|0.7|8.9|49.6|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_sim_Livingroom_Linear_Array_LD10|409|6766|91.9|7.0|1.0|0.8|8.8|50.1|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_sim_Livingroom_Linear_Array_LD11|409|6766|92.1|6.9|1.0|0.7|8.6|51.1|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_sim_Livingroom_Linear_Array_LD12|409|6766|91.9|7.0|1.1|0.8|9.0|50.1|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_sim_Livingroom_Wall_L1C|409|6766|92.2|6.9|0.9|0.8|8.5|49.6|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_sim_Livingroom_Wall_L1L|409|6766|92.0|6.9|1.1|0.7|8.6|50.4|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_sim_Livingroom_Wall_L1R|409|6766|91.9|7.0|1.0|0.8|8.8|50.1|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_sim_Livingroom_Wall_L2L|409|6766|92.2|6.9|0.9|0.8|8.6|49.1|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_sim_Livingroom_Wall_L2R|409|6766|92.0|6.8|1.2|0.7|8.7|49.1|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_sim_Livingroom_Wall_L3L|409|6766|92.1|7.0|0.9|0.8|8.7|48.2|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_sim_Livingroom_Wall_L3R|409|6766|92.7|6.5|0.9|0.8|8.2|46.2|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_sim_Livingroom_Wall_L4L|409|6766|92.2|7.0|0.9|0.7|8.5|48.9|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_sim_Livingroom_Wall_L4R|409|6766|91.8|7.2|1.0|0.8|9.0|52.6|
    
    ### CER
    
    |dataset|Snt|Wrd|Corr|Sub|Del|Ins|Err|S.Err|
    |---|---|---|---|---|---|---|---|---|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_real_Kitchen_Circular_Array_KA1|409|39842|84.9|8.1|7.1|5.4|20.5|76.3|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_real_Kitchen_Circular_Array_KA2|409|39842|84.9|8.2|6.9|5.2|20.2|77.8|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_real_Kitchen_Circular_Array_KA3|409|39842|85.4|8.2|6.5|5.4|20.0|76.5|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_real_Kitchen_Circular_Array_KA4|409|39842|84.9|8.3|6.9|5.3|20.4|76.8|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_real_Kitchen_Circular_Array_KA5|409|39842|84.8|8.3|6.8|5.7|20.9|75.3|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_real_Kitchen_Circular_Array_KA6|409|39842|84.7|8.3|7.0|5.2|20.5|76.5|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_real_Livingroom_Circular_Array_Beam_Circular_Array|409|39842|95.7|2.6|1.7|2.0|6.3|49.9|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_real_Livingroom_Circular_Array_LA1|409|39842|95.0|3.1|1.9|2.6|7.6|54.5|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_real_Livingroom_Circular_Array_LA2|409|39842|94.8|3.2|2.0|2.6|7.8|56.7|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_real_Livingroom_Circular_Array_LA3|409|39842|94.8|3.2|2.1|2.6|7.9|55.7|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_real_Livingroom_Circular_Array_LA4|409|39842|94.7|3.2|2.1|2.7|8.0|56.7|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_real_Livingroom_Circular_Array_LA5|409|39842|95.0|3.2|1.8|2.5|7.5|56.0|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_real_Livingroom_Circular_Array_LA6|409|39842|94.8|3.2|2.0|2.7|7.9|55.3|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_real_Livingroom_Linear_Array_Beam_Linear_Array|409|39842|97.1|1.5|1.4|1.4|4.4|50.6|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_real_Livingroom_Linear_Array_LD02|409|39842|94.8|2.9|2.2|2.7|7.9|59.2|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_real_Livingroom_Linear_Array_LD03|409|39842|95.3|2.8|1.9|2.3|7.0|56.0|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_real_Livingroom_Linear_Array_LD04|409|39842|96.2|2.1|1.7|1.7|5.5|53.8|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_real_Livingroom_Linear_Array_LD05|409|39842|94.8|3.2|2.1|2.5|7.7|59.9|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_real_Livingroom_Linear_Array_LD06|409|39842|95.2|2.8|2.0|2.4|7.1|58.7|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_real_Livingroom_Linear_Array_LD07|409|39842|95.1|2.9|2.0|2.2|7.1|58.2|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_real_Livingroom_Linear_Array_LD08|409|39842|93.9|3.7|2.4|3.3|9.4|58.2|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_real_Livingroom_Linear_Array_LD09|409|39842|96.1|2.1|1.8|1.8|5.7|57.7|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_real_Livingroom_Linear_Array_LD10|409|39842|95.5|2.6|1.9|2.1|6.6|58.7|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_real_Livingroom_Linear_Array_LD11|409|39842|95.7|2.5|1.8|2.0|6.3|55.0|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_real_Livingroom_Linear_Array_LD12|409|39842|95.4|2.7|1.9|2.2|6.8|56.0|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_real_Livingroom_Wall_L1C|409|39842|95.1|2.9|1.9|2.4|7.2|54.8|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_real_Livingroom_Wall_L1L|409|39842|95.3|2.8|1.9|2.3|7.0|53.5|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_real_Livingroom_Wall_L1R|409|39842|95.0|3.0|2.1|2.3|7.3|57.5|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_real_Livingroom_Wall_L2L|409|39842|94.9|3.2|2.0|2.4|7.5|53.8|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_real_Livingroom_Wall_L2R|409|39842|94.4|3.5|2.1|2.8|8.4|57.2|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_real_Livingroom_Wall_L3L|409|39842|95.1|2.9|2.0|2.4|7.3|54.3|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_real_Livingroom_Wall_L3R|409|39842|95.0|3.0|2.0|2.2|7.2|55.3|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_real_Livingroom_Wall_L4L|409|39842|95.2|2.9|1.9|2.2|6.9|55.3|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_real_Livingroom_Wall_L4R|409|39842|95.2|3.0|1.9|2.2|7.0|55.0|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_sim_Kitchen_Circular_Array_KA1|409|39869|90.9|5.8|3.3|3.8|12.9|69.4|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_sim_Kitchen_Circular_Array_KA2|409|39869|90.9|5.7|3.4|3.5|12.7|67.7|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_sim_Kitchen_Circular_Array_KA3|409|39869|91.1|5.5|3.4|3.8|12.7|68.2|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_sim_Kitchen_Circular_Array_KA4|409|39869|90.8|5.9|3.3|4.0|13.2|67.7|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_sim_Kitchen_Circular_Array_KA5|409|39869|90.6|6.0|3.4|4.1|13.5|68.2|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_sim_Kitchen_Circular_Array_KA6|409|39869|91.0|5.7|3.3|4.0|13.0|68.0|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_sim_Livingroom_Circular_Array_Beam_Circular_Array|409|39869|97.5|1.4|1.1|1.1|3.6|46.2|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_sim_Livingroom_Circular_Array_LA1|409|39869|97.0|1.7|1.2|1.2|4.2|48.9|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_sim_Livingroom_Circular_Array_LA2|409|39869|97.0|1.7|1.3|1.3|4.3|50.6|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_sim_Livingroom_Circular_Array_LA3|409|39869|97.1|1.7|1.3|1.2|4.2|49.1|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_sim_Livingroom_Circular_Array_LA4|409|39869|96.7|1.8|1.5|1.4|4.6|52.6|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_sim_Livingroom_Circular_Array_LA5|409|39869|97.0|1.7|1.4|1.2|4.3|49.9|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_sim_Livingroom_Circular_Array_LA6|409|39869|96.9|1.7|1.4|1.3|4.4|49.9|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_sim_Livingroom_Linear_Array_Beam_Linear_Array|409|39869|97.7|1.3|1.0|1.0|3.3|47.4|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_sim_Livingroom_Linear_Array_LD02|409|39869|97.1|1.6|1.3|1.1|4.0|49.1|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_sim_Livingroom_Linear_Array_LD03|409|39869|97.3|1.5|1.2|1.1|3.8|48.2|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_sim_Livingroom_Linear_Array_LD04|409|39869|97.1|1.7|1.2|1.2|4.1|50.4|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_sim_Livingroom_Linear_Array_LD05|409|39869|97.3|1.4|1.3|1.1|3.8|48.9|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_sim_Livingroom_Linear_Array_LD06|409|39869|97.2|1.5|1.3|1.1|4.0|48.9|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_sim_Livingroom_Linear_Array_LD07|409|39869|97.3|1.5|1.2|1.1|3.8|49.1|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_sim_Livingroom_Linear_Array_LD08|409|39869|97.4|1.4|1.2|1.1|3.7|47.9|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_sim_Livingroom_Linear_Array_LD09|409|39869|97.1|1.7|1.3|1.2|4.1|49.6|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_sim_Livingroom_Linear_Array_LD10|409|39869|97.2|1.5|1.3|1.1|4.0|50.1|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_sim_Livingroom_Linear_Array_LD11|409|39869|97.2|1.5|1.3|1.2|4.0|51.1|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_sim_Livingroom_Linear_Array_LD12|409|39869|96.9|1.6|1.4|1.2|4.2|50.1|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_sim_Livingroom_Wall_L1C|409|39869|97.3|1.5|1.2|1.2|3.9|49.6|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_sim_Livingroom_Wall_L1L|409|39869|97.2|1.5|1.3|1.1|4.0|50.4|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_sim_Livingroom_Wall_L1R|409|39869|97.1|1.5|1.3|1.2|4.0|50.1|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_sim_Livingroom_Wall_L2L|409|39869|97.2|1.5|1.3|1.2|4.0|49.1|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_sim_Livingroom_Wall_L2R|409|39869|97.1|1.5|1.4|1.1|4.0|49.1|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_sim_Livingroom_Wall_L3L|409|39869|97.3|1.5|1.2|1.2|4.0|48.2|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_sim_Livingroom_Wall_L3R|409|39869|97.5|1.4|1.1|1.2|3.7|46.2|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_sim_Livingroom_Wall_L4L|409|39869|97.3|1.6|1.2|1.2|3.9|48.9|
    |decode_lm_lm_train_lm_transformer_char_batch_bins350000_accum_grad2_valid.loss.ave_asr_model_valid.acc.ave/dirha_sim_Livingroom_Wall_L4R|409|39869|97.0|1.7|1.3|1.2|4.2|52.6|
  • ASR config
    config: ./conf/tuning/train_asr_transformer_cmvn.yaml
    print_config: false
    log_level: INFO
    dry_run: false
    iterator_type: sequence
    output_dir: exp/asr_train_asr_transformer_cmvn_raw_char_rir_scpdatadirha_irwav.scp_noise_db_range10_17_noise_scpdatadirha_noisewav.scp_speech_volume_normalize1.0_num_workers2_rir_apply_prob1._sp
    ngpu: 1
    seed: 0
    num_workers: 2
    num_att_plot: 3
    dist_backend: nccl
    dist_init_method: env://
    dist_world_size: 4
    dist_rank: 0
    local_rank: 0
    dist_master_addr: localhost
    dist_master_port: 35837
    dist_launcher: null
    multiprocessing_distributed: true
    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
    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
    pretrain_path: null
    init_param: []
    num_iters_per_epoch: null
    batch_size: 20
    valid_batch_size: null
    batch_bins: 16000000
    valid_batch_bins: null
    train_shape_file:
    - exp/asr_stats_tmp_sp/train/speech_shape
    - exp/asr_stats_tmp_sp/train/text_shape.char
    valid_shape_file:
    - exp/asr_stats_tmp_sp/valid/speech_shape
    - exp/asr_stats_tmp_sp/valid/text_shape.char
    batch_type: numel
    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_si284_sp/wav.scp
        - speech
        - kaldi_ark
    -   - dump/raw/train_si284_sp/text
        - text
        - text
    valid_data_path_and_name_and_type:
    -   - dump/raw/dirha_sim_Livingroom_Circular_Array_Beam_Circular_Array/wav.scp
        - speech
        - kaldi_ark
    -   - dump/raw/dirha_sim_Livingroom_Circular_Array_Beam_Circular_Array/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: 0.005
    scheduler: warmuplr
    scheduler_conf:
        warmup_steps: 30000
    token_list:
    - 
    - 
    - 
    - E
    - T
    - A
    - N
    - I
    - O
    - S
    - R
    - H
    - L
    - D
    - C
    - U
    - M
    - P
    - F
    - G
    - Y
    - W
    - B
    - V
    - K
    - .
    - X
    - ''''
    - J
    - Q
    - Z
    - 
    - ','
    - '-'
    - '"'
    - '*'
    - ':'
    - (
    - )
    - '?'
    - '!'
    - '&'
    - ;
    - '1'
    - '2'
    - '0'
    - /
    - $
    - '{'
    - '}'
    - '8'
    - '9'
    - '6'
    - '3'
    - '5'
    - '7'
    - '4'
    - '~'
    - '`'
    - _
    - <*IN*>
    - <*MR.*>
    - \
    - ^
    - 
    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: 0.3
        lsm_weight: 0.1
        length_normalized_loss: false
    use_preprocessor: true
    token_type: char
    bpemodel: null
    non_linguistic_symbols: data/nlsyms.txt
    cleaner: null
    g2p: null
    speech_volume_normalize: 1.0
    rir_scp: data/dirha_ir/wav.scp
    rir_apply_prob: 1.0
    noise_scp: data/dirha_noise/wav.scp
    noise_apply_prob: 1.0
    noise_db_range: '10_17'
    waveaug: null
    waveaug_conf: {}
    frontend: default
    frontend_conf:
        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:
        norm_means: true
        norm_vars: false
        eps: 1.0e-20
    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
    decoder: transformer
    decoder_conf:
        attention_heads: 4
        linear_units: 2028
        num_blocks: 6
        dropout_rate: 0.1
        positional_dropout_rate: 0.1
        self_attention_dropout_rate: 0.0
        src_attention_dropout_rate: 0.0
    required:
    - output_dir
    - token_list
    distributed: true
  • LM config
    config: ./conf/tuning/train_lm_transformer.yaml
    print_config: false
    log_level: INFO
    dry_run: false
    iterator_type: sequence
    output_dir: exp/lm_train_lm_transformer_char_batch_bins350000_accum_grad2
    ngpu: 1
    seed: 0
    num_workers: 1
    num_att_plot: 3
    dist_backend: nccl
    dist_init_method: env://
    dist_world_size: 4
    dist_rank: 0
    local_rank: 0
    dist_master_addr: localhost
    dist_master_port: 55530
    dist_launcher: null
    multiprocessing_distributed: true
    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:
    -   - valid
        - loss
        - min
    keep_nbest_models: 10
    grad_clip: 5.0
    grad_clip_type: 2.0
    grad_noise: false
    accum_grad: 2
    no_forward_run: false
    resume: true
    train_dtype: float32
    use_amp: false
    log_interval: null
    pretrain_path: []
    pretrain_key: []
    num_iters_per_epoch: null
    batch_size: 20
    valid_batch_size: null
    batch_bins: 350000
    valid_batch_bins: null
    train_shape_file:
    - exp/lm_stats_char/train/text_shape.char
    valid_shape_file:
    - exp/lm_stats_char/valid/text_shape.char
    batch_type: numel
    valid_batch_type: null
    fold_length:
    - 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/srctexts
        - text
        - text
    valid_data_path_and_name_and_type:
    -   - dump/raw/test_dev93/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: 0.001
    scheduler: warmuplr
    scheduler_conf:
        warmup_steps: 25000
    token_list:
    - 
    - 
    - 
    - E
    - T
    - A
    - N
    - I
    - O
    - S
    - R
    - H
    - L
    - D
    - C
    - U
    - M
    - P
    - F
    - G
    - Y
    - W
    - B
    - V
    - K
    - .
    - X
    - ''''
    - J
    - Q
    - Z
    - 
    - ','
    - '-'
    - '"'
    - '*'
    - ':'
    - (
    - )
    - '?'
    - '!'
    - '&'
    - ;
    - '1'
    - '2'
    - '0'
    - /
    - $
    - '{'
    - '}'
    - '8'
    - '9'
    - '6'
    - '3'
    - '5'
    - '7'
    - '4'
    - '~'
    - '`'
    - _
    - <*IN*>
    - <*MR.*>
    - \
    - ^
    - 
    init: null
    model_conf:
        ignore_id: 0
    use_preprocessor: true
    token_type: char
    bpemodel: null
    non_linguistic_symbols: data/nlsyms.txt
    cleaner: null
    g2p: null
    lm: transformer
    lm_conf:
        pos_enc: null
        embed_unit: 128
        att_unit: 512
        head: 8
        unit: 2048
        layer: 16
        dropout_rate: 0.1
    required:
    - output_dir
    - token_list
    distributed: true

Files

asr_train_asr_transformer_cmvn_raw_char_rir_scpdatadirha_irwav.scp_noise_db_range10_17_noise_scpdatadirha_noisewav.scp_speech_volume_normalize1.0_num_workers2_rir_apply_prob1._sp_valid.acc.ave.zip

Additional details

Related works

Is supplement to
https://github.com/espnet/espnet (URL)