Automatic Speech Recognition on TALCS (test)
8.22MERREAD
Evaluation results
| Method | Links | ||
|---|---|---|---|
| READCombination Strategy=Segment-level2026.06 | 8.22 | — | |
| Qwen2.5-OmniCategory=Candidate Model2026.06 | 9.21 | — | |
| READCombination Strategy=Sentence-level2026.06 | 9.61 | — | |
| READ (Segment-level combination)SNR=20dB2026.06 | 10.09 | — | |
| Best single-systemSNR=20dB2026.06 | 11.95 | — | |
| OracleDescription=Topline for rescoring2026.06 | 13.07 | — | |
| READ (Segment-level combination)SNR=10dB2026.06 | 13.7 | — | |
| READRescoring Granularity=Sentence-level2026.06 | 14.98 | -20.91 | |
| READCombination Strategy=ROVER integrated with segment-level combination2026.06 | 16.55 | — | |
| Whisper large-v3Category=Candidate Model2026.06 | 16.77 | — | |
| Whisper large-v3Hypothesis Rank=2nd2026.06 | 18.56 | — | |
| Whisper large-v3Hypothesis Rank=3rd2026.06 | 18.57 | — | |
| Whisper large-v3Hypothesis Rank=4th2026.06 | 18.93 | — | |
| Whisper large-v3Hypothesis Rank=1st2026.06 | 18.94 | — | |
| Whisper large-v3Hypothesis Rank=5th2026.06 | 19.69 | — | |
| Whisper mediumCategory=Candidate Model2026.06 | 20.74 | — | |
| ROVERStrategy=ROVER baseline2026.06 | 20.84 | — | |
| Best single-systemSNR=10dB2026.06 | 22.84 | — | |
| ROVERSNR=20dB2026.06 | 23.27 | — | |
| ROVERSNR=10dB2026.06 | 29.08 | — | |
| READ (Segment-level combination)SNR=0dB2026.06 | 30.03 | — | |
| Best single-systemSNR=0dB2026.06 | 41.63 | — | |
| NeMoCategory=Candidate Model2026.06 | 44.47 | — | |
| ROVERSNR=0dB2026.06 | 46.31 | — |