Automatic Speech Recognition on WSJ (eval92)
4.23WER (Initial)Cause-Aware Error Recovery framework
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| Cause-Aware Error Recovery frameworkSetup=Proposed2026.05 | 4.23 | 4.02 | 3.9 | 3.85 | |
| HyPoradiseSetup=HyPoradise (2023), Shot Configuration=Step 1 is 0-shot; Step 3 is 10-shot2026.05 | 7.6 | 7.3 | — | 6.3 |