Speech Separation on WSJ0-2Mix (test)
25.2SDRi (dB)TF-Locoformer (L)
Evaluation Results
| Method | Links | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| TF-Locoformer (L)Domain=TF, #params [M]=22.5, Dynamic Mixing=true2024.08 | 25.2 | 25.1 | — | — | — | — | — | — | — | |
| SepReformer (L) + DM# Params (M)=55.3, GMAC/s (G/s)=155.5, Model scale=L, dynamic mixing (+DM)=true2025.07 | 25.2 | 25.1 | — | — | — | — | — | — | — | |
| TF-Locoformer (M)Domain=TF, #params [M]=15.0, Dynamic Mixing=true2024.08 | 24.7 | 24.6 | — | — | — | — | — | — | — | |
| PRESS-12 @ 12 (M) + FT# Params (M)=22.4, GMAC/s (G/s)=79.7, Model scale=M, Exit point=12, finetuned (+FT)=true2025.07 | 24.55 | 24.36 | — | — | — | — | — | — | — | |
| PRESS-12 @ 12 (M)# Params (M)=22.4, GMAC/s (G/s)=79.7, Model scale=M, Exit point=122025.07 | 24.46 | 24.28 | — | — | — | — | — | — | — | |
| SepReformer-MParams. (M)=17.3, MACs (G/s)=81.32024.06 | 24.4 | 24.2 | — | — | — | — | — | — | — | |
| SepReformer (M)# Params (M)=17.3, GMAC/s (G/s)=81.3, Model scale=M2025.07 | 24.4 | 24.2 | — | — | — | — | — | — | — | |
| PRESS-12 @ 8 (M) + FT# Params (M)=15.6, GMAC/s (G/s)=54.4, Model scale=M, Exit point=8, finetuned (+FT)=true2025.07 | 24.4 | 24.18 | — | — | — | — | — | — | — | |
| TF-Locoformer (L)Domain=TF, #params [M]=22.5, Dynamic Mixing=false2024.08 | 24.3 | 24.2 | — | — | — | — | — | — | — | |
| PRESS-12 @ 8 (M)# Params (M)=15.6, GMAC/s (G/s)=54.4, Model scale=M, Exit point=82025.07 | 24 | 23.47 | — | — | — | — | — | — | — | |
| SepReformer-BParams. (M)=14.2, MACs (G/s)=39.82024.06 | 23.9 | 23.8 | — | — | — | — | — | — | — | |
| SepTDA2Domain=Time, Path=Triple, #params (M)=21.2, Kernel size (L)=12, Data Augmentation=false2024.01 | 23.9 | — | — | 24 | — | — | — | — | — | |
| SepTDA2Domain=T, #params [M]=21.2, Dynamic Mixing=false2024.08 | 23.9 | 24 | — | — | — | — | — | — | — | |
| TF-Locoformer (M)Domain=TF, #params [M]=15.0, Dynamic Mixing=false2024.08 | 23.8 | 23.6 | — | — | — | — | — | — | — | |
| TF-GridNetParams. (M)=14.5, MACs (G/s)=460.82024.06 | 23.6 | 23.5 | — | — | — | — | — | — | — | |
| TF-GridNetDomain=TF, Path=Dual, #params (M)=14.5, Data Augmentation=false2024.01 | 23.6 | — | — | 23.5 | — | — | — | — | — | |
| TF-GridNetDomain=TF, #params [M]=14.4, Dynamic Mixing=false2024.08 | 23.6 | 23.5 | — | — | — | — | — | — | — | |
| PRESS-4 @ 4 (S) + FT# Params (M)=3.4, GMAC/s (G/s)=11.3, Model scale=S, Exit point=4, finetuned (+FT)=true2025.07 | 23.56 | 23.41 | — | — | — | — | — | — | — | |
| SepTDA2Domain=Time, Path=Triple, #params (M)=21.2, Data Augmentation=false2024.01 | 23.5 | — | — | 23.7 | — | — | — | — | — | |
| TF-GridNet (L)# Params=14.4M, GMAC/s=231.12024.10 | 23.5 | — | — | — | 23.4 | — | — | — | — | |
| TF-GridNet (L)# Params (M)=14.4, GMAC/s (G/s)=231.1, Model scale=L2025.07 | 23.5 | 23.4 | — | — | — | — | — | — | — | |
| PRESS-12 @ 4 (M) + FT# Params (M)=8.7, GMAC/s (G/s)=29.1, Model scale=M, Exit point=4, finetuned (+FT)=true2025.07 | 23.43 | 23.27 | — | — | — | — | — | — | — | |
| TF-Gridnet (OA)Params (M)=8.12026.01 | 23.3 | — | — | — | 23.1 | — | — | — | — | |
| SepReformer-SParams. (M)=4.3, MACs (G/s)=21.32024.06 | 23.1 | 23 | — | — | — | — | — | — | — | |
| SPMamba (OA)Params (M)=7.22026.01 | 23.1 | — | — | — | 23 | — | — | — | — | |
| SepReformer (S)# Params (M)=4.5, GMAC/s (G/s)=21.3, Model scale=S2025.07 | 23.1 | 23 | — | — | — | — | — | — | — | |
| PRESS-4 @ 4 (S)# Params (M)=3.4, GMAC/s (G/s)=11.3, Model scale=S, Exit point=42025.07 | 23.08 | 22.91 | — | — | — | — | — | — | — | |
| TF-Locoformer (S)Domain=TF, #params [M]=5.0, Dynamic Mixing=true2024.08 | 23 | 22.8 | — | — | — | — | — | — | — | |
| PRESS-12 @ 4 (M)# Params (M)=8.7, GMAC/s (G/s)=29.1, Model scale=M, Exit point=42025.07 | 22.93 | 22.64 | — | — | — | — | — | — | — | |
| SepMamba (M) + DM# Params=22M, GMAC/s=37, Fw. pass (ms)=27.25, Mem. Usage (GB)=3.042024.10 | 22.9 | 22.7 | — | — | 22.7 | — | — | — | — | |
| SepMamba (M) + DM# Params (M)=22.0, GMAC/s (G/s)=37.0, Model scale=M, dynamic mixing (+DM)=true2025.07 | 22.9 | 22.7 | — | — | — | — | — | — | — | |
| SPMambaParams (M)=6.12026.01 | 22.8 | — | — | — | 22.7 | — | — | — | — | |
| DPMambaParams (M)=15.92026.01 | 22.7 | — | — | — | 22.6 | — | — | — | — | |
| SepReformer-TParams. (M)=3.5, MACs (G/s)=10.42024.06 | 22.6 | 22.4 | — | — | — | — | — | — | — | |
| SepReformer (T)# Params (M)=3.7, GMAC/s (G/s)=10.4, Model scale=T2025.07 | 22.6 | 22.4 | — | — | — | — | — | — | — | |
| ISCITParams. (M)=58.4, MACs (G/s)=252.2, Additional speaker information=true2024.06 | 22.5 | 22.4 | — | — | — | — | — | — | — | |
| SepFormer + DM# Params (M)=26.0, GMAC/s (G/s)=86.9, dynamic mixing (+DM)=true2025.07 | 22.5 | 22.3 | — | — | — | — | — | — | — | |
| SepFormer + DM# Param=26M, Stride=8, Dynamic Mixing (DM)=true2020.10 | 22.4 | 22.3 | — | — | — | — | — | — | — | |
| SepFormerDomain=T, #params [M]=25.7, Dynamic Mixing=true, Speed perturbation=true2024.08 | 22.4 | 22.3 | — | — | — | — | — | — | — | |
| Wavesplit + DM# Param=29M, Stride=1, Dynamic Mixing (DM)=true, uses speaker-ids as additional info=Yes2020.10 | 22.3 | 22.2 | — | — | — | — | — | — | — | |
| WavesplitDomain=T, #params [M]=29.0, Dynamic Mixing=true2024.08 | 22.3 | 22.2 | — | — | — | — | — | — | — | |
| SFSRNetParams. (M)=59.0, MACs (G/s)=124.22024.06 | 22.1 | 22 | — | — | — | — | — | — | — | |
| TF-Locoformer (S)Domain=TF, #params [M]=5.0, Dynamic Mixing=false2024.08 | 22.1 | 22 | — | — | — | — | — | — | — | |
| DPTNetBS=1, L=4s, PT=Speech Enhancement (SE)2020.10 | 21.5 | 21.3 | — | — | — | — | — | — | — | |
| SepEDADomain=Time, Path=Triple, #params (M)=12.5, Data Augmentation=true2024.01 | 21.4 | — | — | 21.2 | — | — | — | — | — | |
| SepMamba (S) + DM# Params=7.2M, GMAC/s=12.46, Fw. pass (ms)=17.84, Mem. Usage (GB)=22024.10 | 21.4 | 21.2 | — | — | 21.2 | — | — | — | — | |
| SepMambaParams (M)=7.22026.01 | 21.4 | — | — | — | 21.2 | — | — | — | — | |
| SepMamba (S) + DM# Params (M)=7.2, GMAC/s (G/s)=12.5, Model scale=S, dynamic mixing (+DM)=true2025.07 | 21.4 | 21.2 | — | — | — | — | — | — | — | |
| TFPSNetDomain=TF, #params [M]=2.7, Dynamic Mixing=false2024.08 | 21.3 | 21.1 | — | — | — | — | — | — | — | |
| TFPSNetParams (M)=2.72026.01 | 21.3 | — | — | — | 21.1 | — | — | — | — | |
| WaveSplitParams. (M)=29.02024.06 | 21.2 | 21 | — | — | — | — | — | — | — | |
| Wavesplit# Param=29M, Stride=1, Dynamic Mixing (DM)=false, uses speaker-ids as additional info=Yes2020.10 | 21.2 | 21 | — | — | — | — | — | — | — | |
| WavesplitDomain=Time, Path=Single, #params (M)=29, Data Augmentation=false2024.01 | 21.2 | — | — | 21 | — | — | — | — | — | |
| WavesplitDomain=T, #params [M]=29.0, Dynamic Mixing=false2024.08 | 21.2 | 21 | — | — | — | — | — | — | — | |
| SandglassetBS=4, PT=from scratch (reported result)2020.10 | 21.2 | 21 | — | — | — | — | — | — | — | |
| SandglassetParams. (M)=2.3, MACs (G/s)=28.82024.06 | 21 | 20.8 | — | — | — | — | — | — | — | |
| DPTNetBS=4, L=4s, PT=Speech Enhancement (SE)2020.10 | 21 | 20.8 | — | — | — | — | — | — | — | |
| DPTNetBS=1, L=4s, PT=from scratch2020.10 | 20.9 | 20.7 | — | — | — | — | — | — | — | |
| S4MParams. (M)=3.6, MACs (G/s)=38.42024.06 | 20.7 | 20.5 | — | — | — | — | — | — | — | |
| S4MParams (M)=3.62026.01 | 20.7 | — | — | — | 20.5 | — | — | — | — | |
| DPTNetModel Size=2.69M2020.07 | 20.6 | — | — | 20.2 | — | — | — | — | — | |
| DPTNet# Param=2.6M, Stride=1, Dynamic Mixing (DM)=false, only SI-SNR and SDR (without improvement) reported=Yes2020.10 | 20.6 | 20.2 | — | — | — | — | — | — | — | |
| DPTNetDomain=Time, Path=Dual, #params (M)=2.7, Data Augmentation=false2024.01 | 20.6 | — | — | 20.2 | — | — | — | — | — | |
| DPTNetDomain=T, #params [M]=2.7, Dynamic Mixing=false2024.08 | 20.6 | 20.2 | — | — | — | — | — | — | — | |
| DPTNetParams (M)=2.72026.01 | 20.6 | — | — | — | 20.2 | — | — | — | — | |
| DPTNetBS=4, PT=from scratch (reported result)2020.10 | 20.6 | 20.2 | — | — | — | — | — | — | — | |
| DPTNetBS=4, L=4s, PT=from scratch2020.10 | 20.6 | 20.4 | — | — | — | — | — | — | — | |
| SepformerParams. (M)=26.0, MACs (G/s)=86.92024.06 | 20.5 | 20.4 | — | — | — | — | — | — | — | |
| GALRD=128, M=4, K=200, Q=8, Size=2.3M, Runtime Memory=730 MiB, GFLOPS=30.82021.01 | 20.5 | 20.3 | — | — | — | — | — | — | — | |
| SepFormer# Param=26M, Stride=8, Dynamic Mixing (DM)=false2020.10 | 20.5 | 20.4 | — | — | — | — | — | — | — | |
| SepFormerDomain=Time, Path=Dual, #params (M)=26, Data Augmentation=false2024.01 | 20.5 | — | — | 20.4 | — | — | — | — | — | |
| SepFormerDomain=T, #params [M]=25.7, Dynamic Mixing=false2024.08 | 20.5 | 20.4 | — | — | — | — | — | — | — | |
| SepFormerParams (M)=26.02026.01 | 20.5 | — | — | — | 20.4 | — | — | — | — | |
| SepFormer# Params (M)=26.0, GMAC/s (G/s)=86.92025.07 | 20.5 | 20.4 | — | — | — | — | — | — | — | |
| VSUNOS# Param=7.5M, Stride=2, Dynamic Mixing (DM)=false2020.10 | 20.4 | 20.1 | — | — | — | — | — | — | — | |
| Gated DPRNNDomain=Time, Path=Dual, #params (M)=7.5, Data Augmentation=false2024.01 | 20.4 | — | — | 20.1 | — | — | — | — | — | |
| DPTNetParams. (M)=2.7, MACs (G/s)=102.52024.06 | 20.3 | 20.2 | — | — | — | — | — | — | — | |
| SudoRM-RFParams (M)=2.72026.01 | 19.1 | — | — | — | 18.9 | — | — | — | — | |
| DPRNNParams. (M)=2.6, MACs (G/s)=88.52024.06 | 19 | 18.8 | — | — | — | — | — | — | — | |
| DPRNN-TasNetnumber of parameters=2.6M2020.05 | 19 | — | — | 18.8 | — | — | — | — | — | |
| DPRNNModel Size=2.6M2020.07 | 19 | — | — | 18.8 | — | — | — | — | — | |
| DualPathRNN# Param=2.6M, Stride=1, Dynamic Mixing (DM)=false2020.10 | 19 | 18.8 | — | — | — | — | — | — | — | |
| DPRNNDomain=Time, Path=Dual, #params (M)=2.6, Data Augmentation=false2024.01 | 19 | — | — | 18.8 | — | — | — | — | — | |
| DPRNNDomain=T, #params [M]=2.6, Dynamic Mixing=false2024.08 | 19 | 18.8 | — | — | — | — | — | — | — | |
| DualPathRNN# Params=2.6M, GMAC/s=42.52, Fw. pass (ms)=101.83, Mem. Usage (GB)=7.312024.10 | 19 | 18.8 | — | — | — | — | — | — | — | |
| DPRNNParams (M)=2.62026.01 | 19 | — | — | — | 18.8 | — | — | — | — | |
| DPRNNBS=4, PT=from scratch (reported result)2020.10 | 19 | 18.8 | — | — | — | — | — | — | — | |
| DualPathRNN# Params (M)=2.6, GMAC/s (G/s)=42.52025.07 | 19 | 18.8 | — | — | — | — | — | — | — | |
| GALRD=128, M=8, K=150, Q=16, Size=2.3M, Runtime Memory=363 MiB, GFLOPS=16.52021.01 | 18.9 | 18.7 | — | — | — | — | — | — | — | |
| DPRNNBS=24, L=2s, PT=Speech Enhancement (SE)2020.10 | 18.9 | 18.6 | — | — | — | — | — | — | — | |
| TDANetParams. (M)=2.3, MACs (G/s)=9.12024.06 | 18.7 | 18.5 | — | — | — | — | — | — | — | |
| TDANetParams (M)=2.32026.01 | 18.7 | — | — | — | 18.5 | — | — | — | — | |
| A-FRCNNParams. (M)=6.1, MACs (G/s)=125.02024.06 | 18.6 | 18.3 | — | — | — | — | — | — | — | |
| A-FRCNNParams (M)=6.12026.01 | 18.6 | — | — | — | 18.3 | — | — | — | — | |
| FurcaNeXtModel Size=51.4M2020.07 | 18.4 | — | — | — | — | — | — | — | — | |
| FurcaNeXt# Param=51.4M, Stride=n.a., Dynamic Mixing (DM)=false2020.10 | 18.4 | — | — | — | — | — | — | — | — | |
| DPRNND=64, M=4, K=200, Size=2.6M, Runtime Memory=929 MiB, GFLOPS=42.32021.01 | 18.1 | 17.9 | — | — | — | — | — | — | — | |
| Deep CASAModel Size=12.8M2020.07 | 18 | — | — | 17.7 | — | — | — | — | — | |
| DeepCASA# Param=12.8M, Stride=1, Dynamic Mixing (DM)=false2020.10 | 18 | 17.7 | — | — | — | — | — | — | — | |
| GALRD=64, M=4, K=200, Q=8, Size=1.5M, Runtime Memory=594 MiB, GFLOPS=21.42021.01 | 17.9 | 17.7 | — | — | — | — | — | — | — |