Table 2 Comparison with other methods on WSJ0-2mix. Best values are marked with bold font
From: Time-domain adaptive attention network for single-channel speech separation
Model | # Param | SI-SNRi (dB) | SDRi (dB) | PESQ | STOI |
|---|---|---|---|---|---|
DPCL++ [32] | 13.6M | 10.8 | - | - | - |
uPIT-BLSTM-ST [12] | 92.7M | - | 10.0 | - | - |
TasNet [17] | - | 13.2 | 13.6 | - | - |
Conv-TasNet [18] | 5.1M | 15.3 | 15.6 | - | - |
DeepCASA [53] | 12.8M | 17.7 | 18.0 | - | - |
FurcaPa [54] | - | - | 18.2 | - | - |
FurcaNeXt [19] | 51.4M | - | 18.4 | - | - |
DPRNN [20] | 2.6M | 18.8 | 19.0 | 3.63 | 0.97 |
SVOICE [55] | 7.5M | 20.1 | 20.4 | - | - |
DPTNet [22] | 2.7M | 20.2 | 20.6 | 3.75 | 0.98 |
SepFormer [23] | 26M | 20.4 | 20.5 | - | - |
TAANet | 5.4M | 20.7 | 20.9 | 3.80 | 0.98 |