Speech Recognition 벤치마크
Speech Recognition on WSJ dev93
Word Error Rate (WER) 낮을수록 좋음
- 2018-12-17 — Convolutional Speech Recognition: Word Error Rate (WER) 6.8
- 2019-04-16 — Convolutional Speech Recognition: Word Error Rate (WER) 6.23
- 2020-05-27 — CTC-CRF VGG-BLSTM: Word Error Rate (WER) 5.7
- 2020-11-11 — CTC-CRF ST-NAS: Word Error Rate (WER) 5.68
| Rank | Model | Word Error Rate (WER) | Paper | Code | Year |
|---|---|---|---|---|---|
| 1 | CTC-CRF ST-NAS | 5.68 | Efficient Neural Architecture Search for End-to-end Speech Recognition via Straight-Through Gradients | thu-spmi/ST-NAS | 2020 |
| 2 | CTC-CRF VGG-BLSTM | 5.7 | CAT: A CTC-CRF based ASR Toolkit Bridging the Hybrid and the End-to-end Approaches towards Data Efficiency and Low Latency | thu-spmi/cat | 2020 |
| 3 | Convolutional Speech Recognition | 6.23 | CRF-based Single-stage Acoustic Modeling with CTC Topology | thu-spmi/cat | 2019 |
| 4 | Convolutional Speech Recognition | 6.8 | Fully Convolutional Speech Recognition | 2018 |