Audio-Visual Speech Recognition on LRS2 50% visual occlusion (test)
6.4WER (Overall)ASR + VSR oracle onb / ocp
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| ASR + VSR oracle onb / ocpInput=A + V2025.12 | 6.4 | 11.7 | 6.6 | 3.5 | 3.6 | |
| DualHyp + RelPromptInput=A + V2025.12 | 13.2 | 20.4 | 16 | 8 | 8.2 | |
| DualHypInput=A + V2025.12 | 14.2 | 21.6 | 17.9 | 8.1 | 9.3 | |
| ASR oracle onb / ocpInput=A2025.12 | 16.7 | 30.9 | 19.4 | 8 | 8.5 | |
| GER w/ Auto-AVSRInput=AV2025.12 | 23.3 | 18.9 | 39 | 17.4 | 18.1 | |
| RobustGERInput=A2025.12 | 24.5 | 39.3 | 33.8 | 11.7 | 13.1 | |
| GERInput=A2025.12 | 24.6 | 39.3 | 34.4 | 11.5 | 13.2 | |
| LipGERInput=AV2025.12 | 24.7 | 39.3 | 34.2 | 12 | 13.4 | |
| Whisper-large-v3Input=A2025.12 | 25.8 | 40 | 36.5 | 12.7 | 14.2 | |
| BRAVEn-largeInput=V2025.12 | 39.7 | — | — | — | — |