--- tags: - espnet - audio - automatic-speech-recognition language: noinfo datasets: - totonac license: cc-by-4.0 --- ## ESPnet2 ASR model ### `espnet/ftshijt_espnet2_asr_totonac_transformer` This model was trained by ftshijt using totonac recipe in [espnet](https://github.com/espnet/espnet/). ### Demo: How to use in ESPnet2 ```bash cd espnet pip install -e . cd els/totonac/asr1 ./run.sh --skip_data_prep false --skip_train true --download_model espnet/ftshijt_espnet2_asr_totonac_transformer ``` # RESULTS ## Environments - date: `Sun Nov 7 09:22:09 EST 2021` - python version: `3.9.7 (default, Sep 16 2021, 13:09:58) [GCC 7.5.0]` - espnet version: `espnet 0.10.4a1` - pytorch version: `pytorch 1.9.0` - Git hash: `` - Commit date: `` ## asr_train_asr_transformer_specaug_raw_bpe250_sp ### WER |dataset|Snt|Wrd|Corr|Sub|Del|Ins|Err|S.Err| |---|---|---|---|---|---|---|---|---| |decode_asr_lm_lm_train_bpe250_valid.loss.ave_asr_model_valid.acc.best/dev|530|3547|59.8|32.9|7.3|6.5|46.7|87.4| |decode_asr_lm_lm_train_bpe250_valid.loss.ave_asr_model_valid.acc.best/test|704|5018|55.5|35.7|8.8|6.1|50.6|92.0| ### CER |dataset|Snt|Wrd|Corr|Sub|Del|Ins|Err|S.Err| |---|---|---|---|---|---|---|---|---| |decode_asr_lm_lm_train_bpe250_valid.loss.ave_asr_model_valid.acc.best/dev|530|22510|88.1|4.4|7.4|3.9|15.8|87.4| |decode_asr_lm_lm_train_bpe250_valid.loss.ave_asr_model_valid.acc.best/test|704|32990|86.9|4.3|8.8|4.0|17.1|92.0| ### TER |dataset|Snt|Wrd|Corr|Sub|Del|Ins|Err|S.Err| |---|---|---|---|---|---|---|---|---| |decode_asr_lm_lm_train_bpe250_valid.loss.ave_asr_model_valid.acc.best/dev|530|9360|70.3|15.8|13.8|4.3|34.0|87.4| |decode_asr_lm_lm_train_bpe250_valid.loss.ave_asr_model_valid.acc.best/test|704|13835|70.5|16.0|13.6|4.4|33.9|92.0| ## ASR config
expand ``` config: conf/tuning/train_asr_transformer_specaug.yaml print_config: false log_level: INFO dry_run: false iterator_type: sequence output_dir: exp/asr_train_asr_transformer_specaug_raw_bpe250_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: 100 patience: 15 val_scheduler_criterion: - valid - loss early_stopping_criterion: - valid - loss - min best_model_criterion: - - valid - acc - max keep_nbest_models: 10 grad_clip: 5 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 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: [] num_iters_per_epoch: null batch_size: 32 valid_batch_size: null batch_bins: 1000000 valid_batch_bins: null train_shape_file: - exp/asr_stats_raw_bpe250_sp/train/speech_shape - exp/asr_stats_raw_bpe250_sp/train/text_shape.bpe valid_shape_file: - exp/asr_stats_raw_bpe250_sp/valid/speech_shape - exp/asr_stats_raw_bpe250_sp/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: - - /tmp/jiatong-7359.okvPvI3Z/raw/train_sp/wav.scp - speech - kaldi_ark - - /tmp/jiatong-7359.okvPvI3Z/raw/train_sp/text - text - text valid_data_path_and_name_and_type: - - /tmp/jiatong-7359.okvPvI3Z/raw/dev/wav.scp - speech - kaldi_ark - - /tmp/jiatong-7359.okvPvI3Z/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: adam optim_conf: lr: 1.0 scheduler: noamlr scheduler_conf: warmup_steps: 4000 token_list: - - - ':' - ▁N - NI - N - ▁IYMA - ▁NA - NA - ▁WA - WA - ▁ - '''' - KA - ▁MA - MA - T - ▁XA - TA - NCHU - WI - ▁LI - ▁NI - PA - YI - ▁PUS - K - ▁PI - ▁X - S - ▁TA - YA - ▁LA - Q - QA - TI - ▁KA - QO - W - ▁KAH - ▁PALA - H - X - XA - ▁KI - A - LH - I - LA - ▁CHA - ▁A - ▁XLI - ▁LHI - U - ▁K - KANI - KU - Y - ▁LU - Á - ▁CHU - O - KI - ▁KIWI - NTLA - ▁TLA - M - ▁TAWA - ▁TI - ▁S - WANI - CHA - LHI - LI - ▁TU - ▁PALHA - Í - ▁CHANÁ - ▁KILHWAMPA - KÁN - ▁WAYMA - E - SA - ▁E - ▁LHU - LHA - PU - ▁LHA - ▁PA - ▁LAK - ▁ANTA - ▁KITI - NCHÚ - SI - TLA - PI - ▁KINI - CHI - ▁PEROH - ▁PU - QÓ - QALHCHIWINA - TU - ▁TLHA - ▁WI - NÁ - ▁KAN - ▁NAYI - CH - 'NO' - ▁U - TSA - MÁ - NQO - ▁ANA - ▁LIKWA - ▁XTA - J - ▁QALH - TO - TÁ - ▁USA - ▁PORQUE - ▁MI - L - ▁TAWÁ - XI - LHAQAPASA - P - CHIWI - WÁ - NTI - ▁JKA - Ú - NTLHA - R - TSI - C - STA - ▁LH - LHU - MPI - ▁I - ▁NILH - ▁KATSI - ▁LHAK - MAKLHAKASKI - ▁WANIKÁN - ▁WIXI - ▁TSI - KÚ - NÍ - ▁PAKS - NU - TLHA - YÁ - KUCHAN - XAQATLI - ▁MAX - ▁LAQAPASA - ▁LAQ - QALH - KATSI - Ó - LAQAPASA - ▁J - ▁QAMA - NTU - MI - KIWI - ▁KIN - ▁XANAT - ▁CHI - JA - ▁IY - ▁TSU - MAKLAKAS - ▁MAQA - LÁ - ▁KATSIYA - ▁TLANKA - ▁STAK - ▁XLA - ▁LHIKWA - ▁SQA - ▁P - TAHNA - ▁TLAQ - ▁JKATSI - MAKLAKASKINKA - YÁW - WATIYA - CHÁ - ▁IPORQUEI - ▁AKXNI - TSU - ▁TSINÓ - ▁STAKA - ▁AKXNÍ - LAKATA - KATSÍ - ▁XALHAK - TLAWAYA - SPUT - ▁XATAWA - QALHCHIWI - PÁ - JU - ▁XAXANAT - ▁PÉREZ - ▁AKTSU - ▁JKI - NTÚ - ▁KATSIYÁ - ▁IESTEI - LAQAPASÁ - ▁MASKI - ▁LAQSQATÁ - ▁TLHANKA - ▁WANIKANI - ▁LÓPEZ - MAKLAKASKINKÁN - ▁ANTÁ - ▁TACHIWÍ - ▁SEBAST - ▁CANO - ▁XKUTNI - ▁UKXILH - TANKAH - LAKASKINQO - LAKAPASTAK - ▁XCHACHAT - TAKAWANÍ - ▁TLÁ - ▁TSINOH - KAXTLAWA - ▁NÚÑEZ - ▁XLAKASKINKA - ▁WÁTIYA - ONCE - Z - É - D - Ñ - V - F - G - '1' - B - init: xavier_uniform input_size: null ctc_conf: dropout_rate: 0.0 ctc_type: builtin reduce: true ignore_nan_grad: true model_conf: ctc_weight: 0.3 lsm_weight: 0.1 length_normalized_loss: false use_preprocessor: true token_type: bpe bpemodel: data/token_list/bpe_unigram250/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: 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: global_mvn normalize_conf: stats_file: exp/asr_stats_raw_bpe250_sp/train/feats_stats.npz preencoder: null preencoder_conf: {} encoder: transformer 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 postencoder: null postencoder_conf: {} decoder: transformer decoder_conf: input_layer: embed num_blocks: 6 linear_units: 2048 dropout_rate: 0.1 required: - output_dir - token_list version: 0.10.4a1 distributed: false ```
### Citing ESPnet ```BibTex @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: ```bibtex @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} } ```