Target Speech Extraction on iEEG Clinical Dataset (test)
13SNROracle Speaker
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| Oracle Speakerrole=Upper Bound2025.02 | 13 | 12.8 | 8.8 | 95.5 | |
| Target Speech Extraction2025.02 | 12.8 | 10.4 | 14.3 | 94.8 | |
| AAD-LLMattention_source=brain-decoded, training_data=additional clinical-15m2025.02 | 12.6 | 11.5 | 11.4 | 94.7 | |
| AAD-LLMattention_source=oracle2025.02 | 12.3 | 10.4 | 13 | 94.3 | |
| AAD-LLMattention_source=brain-decoded2025.02 | 12.2 | 10.3 | 14.7 | 94.1 | |
| BSS + Mel reconstruction2025.02 | 12 | 9.6 | 15.2 | 94.1 | |
| BSS + Envelope reconstruction2025.02 | 11.2 | 7.8 | 20.4 | 90.7 | |
| Blind Speech Separation (BSS)2025.02 | 5.2 | -12.3 | 64.3 | 80.6 | |
| Speech Mixturerole=Input baseline2025.02 | 0.1 | -12.1 | 37.4 | 84.4 |