Voice Separation on WSJ0 3mix
19.5SI-SNRiSepformer
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Sepformer# Param=26M, Dynamic Mixing=true2020.10 | 19.5 | 19.7 | |
| Wavesplit# Param=29M, Dynamic Mixing=true2020.10 | 17.8 | 18.1 | |
| Sepformer# Param=26M, Dynamic Mixing=false2020.10 | 17.6 | 17.9 | |
| Wavesplit# Param=29M, Dynamic Mixing=false2020.10 | 17.3 | 17.6 | |
| MULCAT#params=7.5M2020.02 | 16.9 | — | |
| VSUNOS# Param=7.5M2020.10 | 16.9 | — | |
| DPRNN#params=3.6M, training=by paper authors2020.02 | 14.7 | — | |
| DualPathRNN# Param=2.6M2020.10 | 14.7 | — | |
| IBMtraining=from original paper2020.02 | 12.8 | — | |
| ConvTasNet#params=5.1M, training=from original paper2020.02 | 12.7 | — | |
| ConvTasnet# Param=5.1M2020.10 | 12.7 | 13.1 | |
| IRMtraining=from original paper2020.02 | 12.5 | — | |
| ADANet#params=9.1M, training=from original paper2020.02 | 9.1 | — | |
| DPCL++#params=13.6M, training=from original paper2020.02 | 7.1 | — |