Speech Enhancement on RealMAN (real-recorded)
2.911SIG ScoreDiT-Flow
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| DiT-FlowTrained dataset=StillSonicSet, RTF=0.2302026.03 | 2.911 | 3.684 | 2.402 | 89.6 | |
| SGMSETrained dataset=StillSonicSet, RTF=0.5652026.03 | 2.829 | 3.623 | 2.324 | 88.3 | |
| DiT-FlowTrained dataset=WSJ0+Reverb [13], RTF=0.2302026.03 | 2.754 | 3.413 | 2.184 | 88.5 | |
| StoRMTrained dataset=StillSonicSet, RTF=0.4942026.03 | 2.661 | 3.662 | 2.154 | 88.6 | |
| StoRMTrained dataset=WSJ0+Reverb [13], RTF=0.4942026.03 | 2.481 | 2.976 | 1.884 | 89.1 | |
| SGMSETrained dataset=WSJ0+Reverb [13], RTF=0.5652026.03 | 2.473 | 3.031 | 1.863 | 89.2 | |
| Noisy (lower bound)Trained dataset=N/A, RTF=N/A2026.03 | 1.97 | 2.204 | 1.417 | 86 |