Target Speaker Extraction on ESD (test)
16.67SI-SDRi (dB)EvoTSE
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| EvoTSETrain set=WSJ + ESD, Inference protocol=EvoTSE, k=242026.04 | 16.67 | 3.5 | 17.9 | |
| EvoTSETrain set=WSJ + ESD, Inference protocol=EvoTSE, k=32026.04 | 16.23 | 4.3 | 17.85 | |
| USEF-TFGridNetTrain set=WSJ + ESD, Inference protocol=Standard2026.04 | 13.77 | 9.2 | 17.79 | |
| USEF-TFGridNetTrain set=WSJ + ESD, Inference protocol=Static2026.04 | 13.75 | 9.2 | 17.82 | |
| EvoTSETrain set=WSJ, Inference protocol=EvoTSE, k=242026.04 | 11.34 | 6.4 | 13.59 | |
| EvoTSETrain set=WSJ, Inference protocol=EvoTSE, k=32026.04 | 10.73 | 8.1 | 13.45 | |
| USEF-TFGridNetTrain set=WSJ, Inference protocol=Static2026.04 | 2.09 | 23.9 | 13.37 | |
| USEF-TFGridNetTrain set=WSJ, Inference protocol=Standard2026.04 | 1.81 | 24.3 | 13.28 |