Audio-Visual Speech Recognition on LRS3 (test) (Visual Quality/Robustness Metrics)
43Overall ScoreGER w/ Auto-AVSR
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| GER w/ Auto-AVSRInput=AV2025.12 | 43 | 47.5 | 44.2 | 42 | 38.2 | |
| BRAVEn-largeInput=V2025.12 | 29 | 31.9 | 30.8 | 29.5 | 23.8 | |
| RobustGERInput=A2025.12 | 27.1 | — | — | — | — | |
| GERInput=A2025.12 | 26 | — | — | — | — | |
| LipGERInput=AV2025.12 | 26 | 26.2 | 26 | 25.9 | 26 | |
| Whisper-large-v3Input=A2025.12 | 23.8 | 23.8 | 23.8 | 23.8 | 23.8 | |
| DualHypInput=A + V2025.12 | 10.9 | 12.2 | 10.9 | 10.8 | 9.6 | |
| DualHyp + RelPromptInput=A + V2025.12 | 10.1 | 11 | 10.5 | 10.1 | 8.8 | |
| ASR oracle onb / ocpInput=A2025.12 | 8.3 | — | — | — | — | |
| ASR + VSR oracle onb / ocpInput=A + V2025.12 | 2.9 | 3.2 | 3 | 3 | 2.5 |