Audio-Visual Speech Recognition on LRS2 Speech noise (WER by SNR)
24.13WER (-10 dB)VisG AV-HuBERT
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| VisG AV-HuBERTSize=Large2026.04 | 24.13 | 17.67 | 15.61 | 13.65 | 13.42 | |
| AV-HuBERTSize=Large2026.04 | 27.1 | 19.13 | 15.48 | 14.25 | 13.74 | |
| VisG AV-HuBERTSize=Base2026.04 | 31.09 | 23.74 | 19.07 | 16.71 | 15.25 | |
| AV-HuBERTSize=Base2026.04 | 39.89 | 26.24 | 19.71 | 16.14 | 14.79 |