Automatic Speech Recognition on VCTK noisy
1.52WERREAD
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| READCombination Strategy=ROVER integrated with segment-level combination2026.06 | 1.52 | — | |
| ROVERStrategy=ROVER baseline2026.06 | 1.84 | — | |
| Best single-systemSNR=20dB2026.06 | 2.03 | — | |
| ROVERSNR=20dB2026.06 | 2.09 | — | |
| READCombination Strategy=Segment-level2026.06 | 2.18 | — | |
| READ (Segment-level combination)SNR=20dB2026.06 | 2.2 | — | |
| READCombination Strategy=Sentence-level2026.06 | 2.36 | — | |
| Qwen2.5-OmniCategory=Candidate Model2026.06 | 2.47 | — | |
| READ (Segment-level combination)SNR=10dB2026.06 | 2.68 | — | |
| Best single-systemSNR=10dB2026.06 | 2.71 | — | |
| NeMoCategory=Candidate Model2026.06 | 2.85 | — | |
| ROVERSNR=10dB2026.06 | 3.4 | — | |
| OracleDescription=Topline for rescoring2026.06 | 6.14 | — | |
| READRescoring Granularity=Sentence-level2026.06 | 7.19 | -2.97 | |
| Whisper large-v3Hypothesis Rank=1st2026.06 | 7.41 | — | |
| Whisper large-v3Hypothesis Rank=2nd2026.06 | 8.76 | — | |
| Whisper large-v3Category=Candidate Model2026.06 | 8.87 | — | |
| Whisper large-v3Hypothesis Rank=3rd2026.06 | 9.21 | — | |
| Whisper large-v3Hypothesis Rank=4th2026.06 | 10.85 | — | |
| Whisper large-v3Hypothesis Rank=5th2026.06 | 12.18 | — | |
| READ (Segment-level combination)SNR=0dB2026.06 | 12.77 | — | |
| Best single-systemSNR=0dB2026.06 | 13.28 | — | |
| ROVERSNR=0dB2026.06 | 17.65 | — | |
| Whisper mediumCategory=Candidate Model2026.06 | 18.33 | — |