Audio-Visual Speech Recognition on LRS2 Babble noise (Multi-SNR WER Evaluation)
41.65WER (-10 dB)VisG AV-HuBERT
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| VisG AV-HuBERTSize=Large2026.04 | 41.65 | 28.76 | 19.17 | 15.06 | 13.81 | |
| AV-HuBERTSize=Large2026.04 | 43.06 | 30.18 | 18.59 | 15.57 | 14.38 | |
| AV-HuBERTSize=Base2026.04 | 52.72 | 37.73 | 23.78 | 17.8 | 14.895 | |
| VisG AV-HuBERTSize=Base2026.04 | 53.33 | 36.78 | 23.09 | 17.61 | 15.42 |