Audio-Visual Speech Recognition on LRS2 Babble Noise (Multi-SNR WER)
7.82WER (0 dB)VIB-AVSR
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| VIB-AVSRTraining=Noisy2026.06 | 7.82 | 5.35 | 32.52 | 25.63 | 15.63 | 17.39 | 24.59 | |
| Llama-AVSRTraining=Noisy2026.06 | 8 | 5.74 | 34.87 | 27.55 | 18.09 | 18.85 | 26.84 | |
| VIB-AVSRTraining=Clean2026.06 | 8.55 | 5 | 40.97 | 31.5 | 19.45 | 21.09 | 30.64 | |
| Llama-AVSRTraining=Clean2026.06 | 8.61 | 5.19 | 47.03 | 34.32 | 20.2 | 23.07 | 33.85 |