Speech Enhancement on Valentini 27 (test)
3.4PESQMAE-based universal speech enhancer
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| MAE-based universal speech enhancerPre-train augmentation type=Multi speaker+log1p, Finetune=Y2026.02 | 3.4 | 4.41 | 3.21 | 3.78 | 11.49 | 4.22 | |
| MAE-based universal speech enhancerPre-train augmentation type=Single speaker, Finetune=Y2026.02 | 3.33 | 4.31 | 3.02 | 3.68 | 11.35 | 4.13 | |
| MAE-based universal speech enhancerPre-train augmentation type=Multi speaker, Finetune=Y2026.02 | 3.32 | 4.28 | 2.99 | 3.69 | 11.35 | 4.12 | |
| MAE for restorationPre-train augmentation type=Random TF mask, Finetune=Y2026.02 | 3.25 | 4.28 | 2.95 | 3.51 | 11.34 | 4.28 | |
| MAE-based universal speech enhancerPre-train augmentation type=Multi speaker+log1p, Finetune=N2026.02 | 3.23 | 4.3 | 2.75 | 3.4 | 11.35 | 4.14 | |
| MAE-based universal speech enhancerPre-train augmentation type=Multi speaker, Finetune=N2026.02 | 3.19 | 4.22 | 2.75 | 3.36 | 11.35 | 4.14 | |
| DCUNetPre-train augmentation type=none2026.02 | 3.13 | 4.24 | 4 | 3.69 | — | — | |
| Boosting SSLPre-train augmentation type=none, Finetune=Y2026.02 | 3.05 | 4.4 | 3.52 | 3.74 | — | — | |
| SGMSE+Pre-train augmentation type=none2026.02 | 2.94 | 4.25 | 3.4 | 3.61 | — | 4.56 | |
| UNIVERSEPre-train augmentation type=none2026.02 | 2.9 | 4.03 | 3.11 | 3.46 | — | — | |
| Huang et alPre-train augmentation type=none2026.02 | 2.68 | — | — | — | — | — | |
| VoiceFixerPre-train augmentation type=none2026.02 | 2.43 | — | — | — | — | — | |
| PANN w/ UNetPre-train augmentation type=none2026.02 | 2.28 | 2.43 | 2.96 | 2.3 | — | — | |
| NoisyPre-train augmentation type=none2026.02 | 1.97 | 3.34 | 2.44 | 2.63 | 1.68 | 3.05 |