ASR Error Correction on STOP (test)
5.7WERORACLE
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| ORACLEBackbone=Llama-3.1 (70B), Reference=Ground truth of NE2025.06 | 5.7 | 85.7 | |
| ORACLEBackbone=gpt-4o-mini, Reference=Ground truth of NE2025.06 | 5.7 | 84.7 | |
| DeRAGECBackbone=gpt-4o-mini, Denoising feature=MCQ + PS + Def + Rat2025.06 | 5.8 | 84.2 | |
| DeRAGECBackbone=Llama-3.1 (70B), Denoising feature=MCQ + PS + Def + Rat2025.06 | 5.9 | 83.8 | |
| RAGECBackbone=Llama-3.1 (70B), Denoising feature=PS2025.06 | 6 | 82.8 | |
| RAGECBackbone=gpt-4o-mini, Denoising feature=PS2025.06 | 6 | 82.7 | |
| RAGECBackbone=gpt-4o-mini, Denoising feature=PS + Def2025.06 | 6 | 82.7 | |
| RAGECBackbone=Llama-3.1 (70B), Denoising feature=PS + Def2025.06 | 6.2 | 81.5 | |
| RAGECBackbone=Llama-3.1 (70B), Denoising feature=N (None)2025.06 | 6.5 | 80.7 | |
| RAGECBackbone=Llama-3.1 (70B), Denoising feature=MCQ2025.06 | 6.5 | 80.7 | |
| RAGECBackbone=gpt-4o-mini, Denoising feature=MCQ2025.06 | 6.5 | 81.6 | |
| RAGECBackbone=gpt-4o-mini, Denoising feature=N (None)2025.06 | 6.6 | 81.4 | |
| RAGECBackbone=gpt-4o-mini, Denoising feature=Def2025.06 | 6.9 | 75.2 | |
| RAGECBackbone=Llama-3.1 (70B), Denoising feature=Def2025.06 | 7.2 | 69.7 | |
| GECBackbone=gpt-4o-mini2025.06 | 7.4 | 80.4 | |
| GECBackbone=Llama-3.1 (70B)2025.06 | 7.8 | 80.5 | |
| ASR onlyBackbone=Llama-3.1 (70B)2025.06 | 8.9 | 78.7 | |
| ASR onlyBackbone=gpt-4o-mini2025.06 | 8.9 | 78.7 |