Speech Separation on WHAMR! (test)
17.1ΔSI-SNRTF-GridNet
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| TF-GridNetMACs=228.2G, Params=14.3M2023.10 | 17.1 | — | — | — | — | — | — | |
| MossFormer2Para.(M)=55.72023.12 | 17 | — | — | — | — | — | — | |
| MossFormer(L)Para.(M)=42.1, Data Augmentation=Dynamic Mixing (DM)2023.02 | 16.3 | — | — | — | — | — | — | |
| MossFormer-L + DMMACs=42.8G, Params=42.1M2023.10 | 16.3 | — | — | — | — | — | — | |
| MossFormerPara.(M)=42.12023.12 | 16.3 | — | — | — | — | — | — | |
| MossFormer(M)Para.(M)=25.3, Data Augmentation=Dynamic Mixing (DM)2023.02 | 15.9 | — | — | — | — | — | — | |
| TD-Conformer-XL+DMSubsampling Factor (P)=64, MACs=63.6G, Params=102.2M2023.10 | 14.6 | — | — | — | — | — | — | |
| QDPN + DMParams=200M2023.10 | 14.4 | — | — | — | — | — | — | |
| QDPNPara.(M)=200.02023.12 | 14.4 | — | — | — | — | — | — | |
| TD-Conformer-XL+DMSubsampling Factor (P)=125, MACs=63.9G, Params=102.7M2023.10 | 14.3 | — | — | — | — | — | — | |
| SepFormerPara.(M)=25.7, Data Augmentation=Dynamic Mixing (DM)2023.02 | 14 | — | — | — | — | — | — | |
| SepFormerDynamic Mixing=true2022.02 | 14 | 13 | — | — | — | — | — | |
| SepFormer + DMMACs=60.7G, Params=25.6M2023.10 | 14 | — | — | — | — | — | — | |
| SepFormerPara.(M)=25.72023.12 | 14 | — | — | — | — | — | — | |
| SepFormerSpeed Augmentation=true2022.02 | 13.7 | 12.7 | — | — | — | — | — | |
| TD-Conformer-L+DMSubsampling Factor (P)=64, MACs=21.9G, Params=25.9M2023.10 | 13.4 | — | — | — | — | — | — | |
| WavesplitData augmentation=true2021.10 | 13.2 | 12.2 | — | — | — | — | — | |
| WavesplitPara.(M)=29, Data Augmentation=Dynamic Mixing (DM)2023.02 | 13.2 | — | — | — | — | — | — | |
| WavesplitDynamic Mixing=true2022.02 | 13.2 | 12.2 | — | — | — | — | — | |
| TD-Conformer-L+DMSubsampling Factor (P)=125, MACs=22.0G, Params=26.2M2023.10 | 13.2 | — | — | — | — | — | — | |
| WavesplitPara.(M)=292023.12 | 13.2 | — | — | — | — | — | — | |
| QDPNParams=200M2023.10 | 13.1 | — | — | — | — | — | — | |
| TD-Conformer-XLSubsampling Factor (P)=64, MACs=63.6G, Params=102.2M2023.10 | 13.1 | — | — | — | — | — | — | |
| TD-Conformer-XLSubsampling Factor (P)=125, MACs=63.9G, Params=102.7M2023.10 | 13 | — | — | — | — | — | — | |
| TD-Conformer-LSubsampling Factor (P)=125, MACs=22.0G, Params=26.2M2023.10 | 12.5 | — | — | — | — | — | — | |
| DPRNN-SRSSN2021.10 | 12.3 | 11.4 | — | — | — | — | — | |
| DPTNET-SRSSN2021.10 | 12.3 | 11.3 | — | — | — | — | — | |
| TD-Conformer-LSubsampling Factor (P)=64, MACs=21.9G, Params=25.9M2023.10 | 12.3 | — | — | — | — | — | — | |
| Nachmani et al.2021.10 | 12.2 | — | — | — | — | — | — | |
| VSUNOSPara.(M)=7.52023.02 | 12.2 | — | — | — | — | — | — | |
| VSUNOSPara.(M)=7.52023.12 | 12.2 | — | — | — | — | — | — | |
| DPTNET-TASNET2021.10 | 12.1 | 11.1 | — | — | — | — | — | |
| WavesplitData augmentation=false2021.10 | 12 | 11.1 | — | — | — | — | — | |
| TD-Conformer-M+DMSubsampling Factor (P)=64, MACs=8.5G, Params=6.7M2023.10 | 12 | — | — | — | — | — | — | |
| TD-Conformer-M+DMSubsampling Factor (P)=125, MACs=8.6G, Params=6.8M2023.10 | 11.9 | — | — | — | — | — | — | |
| TD-Conformer-MSubsampling Factor (P)=64, MACs=8.5G, Params=6.7M2023.10 | 11.7 | — | — | — | — | — | — | |
| TD-Conformer-MSubsampling Factor (P)=125, MACs=8.6G, Params=6.8M2023.10 | 11.6 | — | — | — | — | — | — | |
| SepFormerMACs=60.7G, Params=25.6M2023.10 | 11.5 | — | — | — | — | — | — | |
| SepFormer2022.02 | 11.4 | 10.4 | — | — | — | — | — | |
| DTCN + DMMACs=3.7G, Params=3.6M2023.10 | 11.1 | — | — | — | — | — | — | |
| Cascaded-Bi-LSTM-TASNET2021.10 | 10.8 | — | — | — | — | — | — | |
| TD-Conformer-SSubsampling Factor (P)=64, MACs=3.7G, Params=1.8M2023.10 | 10.5 | — | — | — | — | — | — | |
| TD-Conformer-SSubsampling Factor (P)=125, MACs=3.7G, Params=1.8M2023.10 | 10.5 | — | — | — | — | — | — | |
| DPRNN-TASNET2021.10 | 10.3 | — | — | — | — | — | — | |
| DPRNNPara.(M)=2.62023.02 | 10.3 | — | — | — | — | — | — | |
| DPRNNPara.(M)=2.62023.12 | 10.3 | — | — | — | — | — | — | |
| DTCNMACs=3.7G, Params=3.6M2023.10 | 10.2 | — | — | — | — | — | — | |
| Conv-TasNetMACs=3.6G, Params=5.1M2023.10 | 9.7 | — | — | — | — | — | — | |
| TD-Conformer-S+DMSubsampling Factor (P)=64, MACs=3.7G, Params=1.8M2023.10 | 9.7 | — | — | — | — | — | — | |
| TD-Conformer-S+DMSubsampling Factor (P)=125, MACs=3.7G, Params=1.8M2023.10 | 9.7 | — | — | — | — | — | — | |
| Bi-LSTM-TASNET2021.10 | 9.2 | — | — | — | — | — | — | |
| BiLSTM Tasnet2022.02 | 9.2 | — | — | — | — | — | — | |
| FlexIO + SIPSnoise scale κ=02026.05 | 8.51 | — | 1.56 | 3.68 | 4.01 | 3.01 | 23.43 | |
| FlexIO2026.05 | 8.45 | — | 1.76 | 3.62 | 3.54 | 2.79 | 21.5 | |
| Conv-TASNET2021.10 | 8.3 | — | — | — | — | — | — | |
| Conv-TasNetPara.(M)=5.12023.02 | 8.3 | — | — | — | — | — | — | |
| Conv-TasNet2022.02 | 8.3 | — | — | — | — | — | — | |
| Conv-TasNetPara.(M)=5.12023.12 | 8.3 | — | — | — | — | — | — | |
| Mixture2026.05 | 7.2 | — | 1.08 | 2.53 | 1.21 | 1.35 | 92.62 | |
| SepFormer2026.05 | 6.99 | — | 1.55 | 3.13 | 2.06 | 2.34 | 32.22 | |
| SepFormer + SIPSnoise scale κ=02026.05 | 6.79 | — | 1.41 | 3.36 | 3.01 | 2.6 | 34.29 | |
| STFT-ConformerMACs=1.8G, Params=57.5M2023.10 | 6.7 | — | — | — | — | — | — |