Speech Recognition On Wenetspeech
評価指標
Character Error Rate (CER)
評価結果
このベンチマークにおける各モデルのパフォーマンス結果
モデル名 | Character Error Rate (CER) | Paper Title | Repository |
|---|---|---|---|
| Kaldi | 9.07 | WenetSpeech: A 10000+ Hours Multi-domain Mandarin Corpus for Speech Recognition | |
| Conformer-MoE (32e) | 7.49 | 3M: Multi-loss, Multi-path and Multi-level Neural Networks for speech recognition | |
| Paraformer-large | 6.97 | FunASR: A Fundamental End-to-End Speech Recognition Toolkit | |
| Conformer-MoE (16e) | 7.67 | 3M: Multi-loss, Multi-path and Multi-level Neural Networks for speech recognition | |
| Wenet | 8.88 | WenetSpeech: A 10000+ Hours Multi-domain Mandarin Corpus for Speech Recognition | |
| Espnet | 9.7 | WenetSpeech: A 10000+ Hours Multi-domain Mandarin Corpus for Speech Recognition | |
| Zipformer+pruned transducer (no external language model) | 7.29 | Zipformer: A faster and better encoder for automatic speech recognition | |
| Conformer-MoE (64e) | 7.19 | 3M: Multi-loss, Multi-path and Multi-level Neural Networks for speech recognition |
0 of 8 row(s) selected.