Audio-Visual Speech Recognition on LRS2 Speech Noise (SNR Breakdown)
5.08Error Rate (0 dB SNR)VIB-AVSR
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| VIB-AVSRTraining=Noisy2026.06 | 5.08 | 3.98 | 27.52 | 13.23 | 8.9 | 11.74 | 16.55 | |
| Llama-AVSRTraining=Noisy2026.06 | 5.21 | 3.84 | 27.66 | 14.94 | 9.14 | 12.16 | 17.24 | |
| Llama-AVSRTraining=Clean2026.06 | 5.3 | 4.32 | 42.13 | 18.44 | 10.78 | 16.2 | 23.78 | |
| VIB-AVSRTraining=Clean2026.06 | 5.36 | 4.39 | 37.44 | 17.37 | 10.69 | 15.05 | 21.83 |