Speech Enhancement on VoiceBank-DEMAND
3.61PESQPrimeK-Net
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| PrimeK-NetYear=2025, Params=1.41M2026.06 | 3.61 | — | 4.81 | 3.98 | 4.35 | — | |
| SE-MambaParams (M)=2.262026.06 | 3.55 | 0.96 | 4.77 | 3.95 | 4.26 | — | |
| MP-SENetParams (M)=2.052026.06 | 3.5 | 0.96 | 4.73 | 3.95 | 4.22 | — | |
| QC-GAN (Base)Params (M)=0.892026.06 | 3.48 | 0.95 | 4.6 | 3.65 | 4.1 | — | |
| LaCo-SENetL_enc=15, L_dec=15, Year=2026, τ (ms)=200.0, Params=1.37M, Padding=Symmetric2026.06 | 3.47 | 0.957 | 4.69 | 3.79 | 4.17 | — | |
| LaCo-SENetL_enc=5, L_dec=5, Year=2026, τ (ms)=75.0, Params=1.37M2026.06 | 3.43 | 0.954 | 4.66 | 3.78 | 4.12 | — | |
| CMGANAttention Type=Standard, Params (M)=1.832026.05 | 3.41 | 0.96 | 4.63 | 3.94 | 4.12 | — | |
| CMGANParams (M)=1.832026.06 | 3.41 | 0.96 | 4.63 | 3.94 | 4.12 | — | |
| LaCo-SENetL_enc=3, L_dec=3, Year=2026, τ (ms)=50.0, Params=1.37M2026.06 | 3.4 | 0.953 | 4.63 | 3.72 | 4.09 | — | |
| LaCo-SENetL_enc=1, L_dec=1, Year=2026, τ (ms)=25.0, Params=1.37M2026.06 | 3.36 | 0.953 | 4.62 | 3.72 | 4.07 | — | |
| LaCo-SENetL_enc=0, L_dec=0, Year=2026, τ (ms)=12.5, Params=1.37M2026.06 | 3.35 | 0.952 | 4.61 | 3.71 | 4.05 | — | |
| DPT-FSNetParams (M)=0.882026.06 | 3.33 | 0.96 | 4.58 | 3.72 | 4 | — | |
| SEMambaYear=2024, Params=1.41M2026.06 | 3.29 | 0.95 | — | — | — | — | |
| aTENNuateYear=2025, τ (ms)=46.5, Params=0.84M2026.06 | 3.27 | — | 4.57 | 2.85 | 3.96 | — | |
| xLSTM-SENetYear=2025, Params=∼2.2M2026.06 | 3.26 | 0.95 | 4.57 | 3.79 | 4 | — | |
| Real-valued ConformerAttention Type=Standard, Params (M)=3.192026.05 | 3.25 | 0.95 | 4.43 | 3.68 | 3.88 | 19.09 | |
| MANNERDomain=T2022.03 | 3.21 | 95 | 4.53 | 3.65 | 3.91 | — | |
| NSE-CATNetAttention Type=Standard, Params (M)=3.572026.05 | 3.19 | 0.96 | 4.41 | 3.66 | 3.82 | — | |
| QConformerAttention Type=Shared Score, Params (M)=0.802026.05 | 3.18 | 0.95 | 4.36 | 3.65 | 3.79 | 19.36 | |
| DeepFilterNet3Attention Type=–, Params (M)=–2026.05 | 3.17 | 0.94 | 4.34 | 3.61 | 3.77 | — | |
| DFNet3Year=2023, τ (ms)=40, Params=2.13M2026.06 | 3.17 | 0.944 | 4.34 | 3.61 | 3.77 | — | |
| MetricGAN+Domain=T-F2022.03 | 3.15 | — | 4.14 | 3.16 | 3.64 | — | |
| MetricGAN+2022.02 | 3.15 | — | — | — | — | — | |
| MetricGAN+Attention Type=–, Params (M)=–2026.05 | 3.15 | 0.94 | 4.14 | 3.18 | 3.64 | — | |
| MetricGAN+2026.06 | 3.15 | — | 4.14 | 3.16 | 3.64 | — | |
| SE-ConformerAttention Type=Standard, Params (M)=–2026.05 | 3.13 | 0.95 | 4.45 | 3.55 | 3.82 | — | |
| SN-NetDomain=T-F2022.03 | 3.12 | — | 4.39 | 3.6 | 3.77 | — | |
| MANNER (small)Domain=T2022.03 | 3.12 | 95 | 4.45 | 3.61 | 3.82 | — | |
| AIATversion=Magnitude2022.02 | 3.11 | 0.949 | — | — | — | — | |
| QConformerAttention Type=Hamilton, Params (M)=0.802026.05 | 3.11 | 0.95 | 4.3 | 3.64 | 3.73 | 19.64 | |
| DEMUCS (large)Domain=T2022.03 | 3.07 | 95 | 4.31 | 3.4 | 3.63 | — | |
| QTransformerAttention Type=Hamilton, Params (M)=0.622026.05 | 3.07 | 0.95 | 4.26 | 3.61 | 3.68 | 19.62 | |
| QTransformerAttention Type=Shared Score, Params (M)=0.622026.05 | 3.07 | 0.95 | 4.3 | 3.62 | 3.71 | 19.69 | |
| SepFormer2022.02 | 3.03 | 0.949 | — | — | — | — | |
| UNIVERSE++Attention Type=–, Params (M)=107.52026.05 | 3.02 | 0.86 | — | — | — | 18.62 | |
| PHASENDomain=T-F2022.03 | 2.99 | — | 4.21 | 3.55 | 3.62 | — | |
| YinPHASEN2022.02 | 2.99 | — | — | — | — | — | |
| TSTNNDomain=T2022.03 | 2.96 | 95 | 4.33 | 3.53 | 3.67 | — | |
| TSTNNAttention Type=–, Params (M)=0.922026.05 | 2.96 | 0.95 | 4.1 | 3.77 | 3.52 | — | |
| DeepMMSE2022.02 | 2.95 | — | — | — | — | — | |
| GaGNetYear=2022, τ (ms)=∼10, Params=5.94M2026.06 | 2.94 | — | 4.26 | 3.45 | 3.59 | — | |
| CRGAN2022.02 | 2.92 | 0.94 | — | — | — | — | |
| MetricGAN2022.02 | 2.86 | — | — | — | — | — | |
| QDenseNetAttention Type=– (No Bottleneck), Params (M)=0.742026.05 | 2.8 | 0.94 | 3.84 | 3.41 | 3.32 | 18.42 | |
| QTNAttention Type=QConv-based attention, Params (M)=0.632026.05 | 2.76 | 0.94 | 3.76 | 3.43 | 3.26 | 19.55 | |
| CNNTransformer2022.02 | 2.65 | 0.915 | — | — | — | — | |
| DCCRNAttention Type=–, Params (M)=3.072026.05 | 2.59 | 0.94 | — | — | — | 18.74 | |
| Wave U-NetDomain=T2022.03 | 2.4 | — | 3.52 | 3.24 | 2.96 | — | |
| RNNoiseYear=2018, τ (ms)=10, Params=0.06M2026.06 | 2.33 | 0.922 | 3.4 | 2.51 | 2.84 | — | |
| SEGANDomain=T2022.03 | 2.16 | — | 3.48 | 2.94 | 2.8 | — | |
| SEGANAttention Type=–, Params (M)=97.472026.05 | 2.16 | 0.92 | 3.48 | 2.94 | 2.8 | — | |
| SEGANParams (M)=97.472026.06 | 2.16 | 0.92 | 3.48 | 2.94 | 2.8 | — | |
| NoisyAttention Type=–, Params (M)=–2026.05 | 1.97 | 0.91 | 3.35 | 2.44 | 2.63 | — | |
| Noisy2026.06 | 1.97 | 0.91 | 3.35 | 2.44 | 2.63 | — | |
| Noisy2026.06 | 1.97 | 0.921 | 3.35 | 2.44 | 2.63 | — |