Contextual Automatic Speech Recognition on Aishell-1
3.84CERMulti-grained
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| Multi-grainedTop-K=10, Retrieval=JSPG2026.05 | 3.84 | 2.7 | 3.96 | 95.53 | |
| Multi-grainedTop-K=50, Retrieval=JSPG2026.05 | 3.86 | 3.1 | 3.94 | 94.42 | |
| Multi-grainedTop-K=100, Retrieval=JSPG2026.05 | 3.9 | 3.35 | 3.96 | 94.06 | |
| Multi-grainedTop-K=1, Retrieval=JSPG2026.05 | 3.97 | 4.9 | 3.87 | 89.43 | |
| LLM-RefineTop-K=10, Retrieval=JSPG2026.05 | 4.16 | 3.3 | 4.25 | 93.23 | |
| LLM-RefineTop-K=50, Retrieval=JSPG2026.05 | 4.19 | 3.9 | 4.22 | 91.55 | |
| LLM-RefineTop-K=100, Retrieval=JSPG2026.05 | 4.27 | 4.4 | 4.26 | 90.8 | |
| LLM-RefineTop-K=1, Retrieval=JSPG2026.05 | 4.45 | 7 | 4.18 | 85.77 | |
| CopyNETop-K=10, Retrieval=JSPG2026.05 | 4.7 | 3.9 | 4.78 | 93.51 | |
| CopyNETop-K=50, Retrieval=JSPG2026.05 | 4.74 | 4.2 | 4.8 | 92.8 | |
| CopyNETop-K=100, Retrieval=JSPG2026.05 | 4.78 | 4.35 | 4.82 | 92.53 | |
| CopyNETop-K=1, Retrieval=JSPG2026.05 | 4.8 | 5.6 | 4.72 | 88.19 | |
| Whisper BaseModel=Baseline2026.05 | 5.2 | 10.4 | 4.7 | 80.6 |