ESPnet2 pretrained model, kamo-naoyuki/hkust_asr_train_asr_transformer2_raw_zh_char_batch_bins20000000_ctc_confignore_nan_gradtrue_sp_valid.acc.ave, fs=16k, lang=zh

作者：kamo-naoyuki · 发表于：Zenodo (CERN European Organization for Nuclear Research) · 年份：2021 · DOI：10.5281/zenodo.4430974 · 研究领域：Power Systems and Technologies

This model was trained by kamo-naoyuki using hkust recipe in espnet. <strong>Python API</strong><pre><code class="language-python">See https://github.com/espnet/espnet_model_zoo</code></pre> <strong>Evaluate in the recipe</strong><pre><code class="language-bash">git clone https://github.com/espnet/espnet cd espnet git checkout db7dfea809ef919332bd28c392b78378a8df5a77 pip install -e . cd egs2/hkust/asr1 ./run.sh --skip_data_prep false --skip_train true --download_model kamo-naoyuki/hkust_asr_train_asr_transformer2_raw_zh_char_batch_bins20000000_ctc_confignore_nan_gradtrue_sp_valid.acc.ave</code> </pre> <strong>Results</strong><pre><code> # RESULTS ## Environments - date: `Mon Jan 11 08:44:15 JST 2021` - python version: `3.8.5 (default, Sep 4 2020, 07:30:14) [GCC 7.3.0]` - espnet version: `espnet 0.9.6` - pytorch version: `pytorch 1.4.0` - Git hash: `db7dfea809ef919332bd28c392b78378a8df5a77` - Commit date: `Mon Jan 4 09:44:09 2021 +0900` ## asr_train_asr_transformer2_raw_zh_char_batch_bins20000000_ctc_confignore_nan_gradtrue_sp ### WER |dataset|Snt|Wrd|Corr|Sub|Del|Ins|Err|S.Err| |---|---|---|---|---|---|---|---|---| |decode_lm_lm_train_lm_transformer_zh_char_optim_conflr0.001_scheduler_confwarmup_steps25000_batch_bins3000000_accum_grad1_valid.loss.ave_asr_model_valid.acc.ave/dev|5413|5240|30.7|69.1|0.2|0.4|69.7|67.4| |decode_lm_lm_train_lm_transformer_zh_char_optim_conflr0.001_scheduler_confwarmup_steps25000_batch_bins3000000_accum_grad8_valid.loss.ave_asr_model_valid.acc.ave/...