Audio-Visual Speech Recognition on Simulated LRS2 19 (test)
5.2WER (-5 dB)UBG-Net
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| UBG-NetNumber of interfering speakers (Int.)=12026.07 | 5.2 | 3.1 | 3.1 | 2.6 | — | 3.9 | |
| BaselineNumber of interfering speakers (Int.)=12026.07 | 6.4 | 3.5 | 3.4 | 2.8 | — | 4.1 | |
| UBG-NetNumber of interfering speakers (Int.)=22026.07 | 8.3 | 4.1 | 3.2 | 3 | — | — | |
| BaselineNumber of interfering speakers (Int.)=22026.07 | 9 | 4.4 | 3.2 | 2.8 | — | — | |
| Qwen3-OmniNumber of interfering speakers (Int.)=12026.07 | 96.4 | 30 | 13.5 | 5.6 | — | 32.8 | |
| WhisperNumber of interfering speakers (Int.)=12026.07 | 96.6 | 30.4 | 14.1 | 6 | — | 33.8 | |
| Qwen3-OmniNumber of interfering speakers (Int.)=22026.07 | 100 | 30.1 | 10.8 | 4.8 | — | — | |
| WhisperNumber of interfering speakers (Int.)=22026.07 | 104 | 30.4 | 13.1 | 5.4 | — | — | |
| BaselineNumber of interfering speakers (Int.)=02026.07 | — | — | — | — | 2.1 | — | |
| Qwen3-OmniNumber of interfering speakers (Int.)=02026.07 | — | — | — | — | 4.2 | — | |
| UBG-NetNumber of interfering speakers (Int.)=02026.07 | — | — | — | — | 2.2 | — | |
| WhisperNumber of interfering speakers (Int.)=02026.07 | — | — | — | — | 4 | — |