Speech Recognition on L2-Arctic
15.54Arabic WEREM-tok-b
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| EM-tok-bgranularity=token-level, beam-search=true2026.05 | 15.54 | 18.84 | 8.63 | 11.86 | 15.19 | 27.17 | 16.21 | |
| EM-tokgranularity=token-level2026.05 | 16.21 | 18.83 | 9.3 | 12.45 | 17.57 | 27.95 | 17.05 | |
| EM-seqgranularity=sequence-level2026.05 | 16.82 | 20.23 | 9.52 | 12.81 | 18.04 | 28.67 | 17.68 | |
| Greedy-EM2026.05 | 18.45 | 19.67 | 10.27 | 14.23 | 19.47 | 30.53 | 18.77 | |
| Source2026.05 | 19.22 | 22.71 | 10.39 | 14.56 | 18.31 | 30.89 | 19.35 | |
| ASR-TRABackbone=Whisper-Tiny2026.03 | 22.92 | 25.14 | 14.18 | 13.76 | 39.44 | 53.84 | 28.21 | |
| BaselineBackbone=Whisper-Tiny2026.03 | 32.74 | 28.03 | 14.62 | 13.42 | 42.64 | 61.21 | 32.11 | |
| SGEMBackbone=Whisper-Tiny2026.03 | 35.19 | 28.03 | 12.61 | 12.12 | 41.78 | 58.68 | 31.4 | |
| SUTABackbone=Whisper-Tiny2026.03 | 35.94 | 27.68 | 14.32 | 13.56 | 41.25 | 62.78 | 32.59 |