Audio-Visual Speech Recognition on LRS2 Music noise (WER Multi-SNR)
22.76WER (-10 dB SNR)VisG AV-HuBERT
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| VisG AV-HuBERTSize=Large2026.04 | 22.76 | 17.34 | 14.53 | 13.72 | 13.24 | |
| AV-HuBERTSize=Large2026.04 | 23.6 | 17.58 | 15.07 | 14.14 | 14.02 | |
| VisG AV-HuBERTSize=Base2026.04 | 30.07 | 22.1 | 17.34 | 15.3 | 14.32 | |
| AV-HuBERTSize=Base2026.04 | 30.5 | 22.26 | 17.69 | 15.26 | 14.11 |