Speech Separation on WSJ0-3mix (clean)
19.5Delta SI-SNR (dB)SepFormer
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| SepFormerModel size=26M, Domain=Learnable latent domain-based, Data Augmentation=true2021.10 | 19.5 | 19.7 | |
| DPTNET-SRSSNModel size=5.7M, Domain=Learnable latent domain-based2021.10 | 19.4 | 19.6 | |
| DPRNN-SRSSNModel size=7.5M, Domain=Learnable latent domain-based2021.10 | 18.8 | 19 | |
| WavesplitModel size=29M, Domain=Learnable latent domain-based, Data Augmentation=true2021.10 | 17.8 | 18.1 | |
| SepFormerModel size=26M, Domain=Learnable latent domain-based2021.10 | 17.6 | 17.9 | |
| WavesplitModel size=29M, Domain=Learnable latent domain-based2021.10 | 17.3 | 17.6 | |
| Nachmani et al.Model size=7.5M, Domain=Learnable latent domain-based2021.10 | 16.9 | — | |
| DPTNET-TASNETModel size=2.7M, Domain=Learnable latent domain-based2021.10 | 16.2 | 16.5 | |
| DPRNN-TASNETModel size=2.6M, Domain=Learnable latent domain-based2021.10 | 15.7 | 16 | |
| Conv-TASNETModel size=5.1M, Domain=Learnable latent domain-based2021.10 | 12.7 | 13.1 | |
| E2EPEDomain=Learnable latent domain-based2021.10 | 12.5 | 13 | |
| DPCL++Model size=13.6M, Domain=Frequency domain-based2021.10 | 7.1 | — | |
| uPIT-Bi-LSTM-STModel size=92.7M, Domain=Frequency domain-based2021.10 | — | 7.7 |