Audio-Visual Speech Recognition on LRS2 Random noise (WER by SNR)
22.21WER (-10 dB)VisG AV-HuBERT
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| VisG AV-HuBERTSize=Large2026.04 | 22.21 | 17.1 | 14.91 | 13.65 | 13.33 | |
| AV-HuBERTSize=Large2026.04 | 23.42 | 17.67 | 14.88 | 14.5 | 14.43 | |
| AV-HuBERTSize=Base2026.04 | 28.21 | 21.45 | 17.14 | 15.25 | 14.31 | |
| VisG AV-HuBERTSize=Base2026.04 | 29.38 | 21.36 | 17.21 | 15.46 | 14.32 |