Audio-Visual Speech Recognition on LRS3 All Samples (test)
3.7WER (10 dB)USR 2.0
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| USR 2.0Beam size=30, Zero-shot evaluation=true, Noise augmentation=None, Labelled training data=30 hours2026.02 | 3.7 | 5.6 | 14 | 33.1 | 14.1 | |
| USRBeam size=30, Zero-shot evaluation=true, Noise augmentation=None, Labelled training data=30 hours2026.02 | 4 | 6.4 | 14.8 | 36.5 | 15.4 | |
| AV-HuBERTBeam size=30, Zero-shot evaluation=true, Noise augmentation=None, Labelled training data=30 hours2026.02 | 5.8 | 10.5 | 24.3 | 56.9 | 24.4 | |
| BRAVEnBeam size=30, Zero-shot evaluation=true, Noise augmentation=None, Labelled training data=30 hours2026.02 | 7.2 | 14 | 32.2 | 54.6 | 27 |