Speech Enhancement on DNS Challenge Real Recordings (test)
4.206SIG ScoreMAGE
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| MAGEctf=true2025.09 | 4.206 | 4.145 | 3.787 | — | — | |
| MAGEctf=true, corrector=true2025.09 | 4.191 | 3.924 | 3.666 | — | — | |
| MAGEcorrector=true2025.09 | 4.098 | 4.309 | 3.744 | — | — | |
| MAGE2025.09 | 3.83 | 4.302 | 3.5 | — | — | |
| FlowSE2025.09 | 3.643 | 4.1 | 3.271 | — | — | |
| SELMSystem Architecture/Backbone=SELM [25]2026.04 | 3.591 | 3.435 | 3.124 | — | — | |
| SELMModel Category=VQ-Based Models2026.06 | 3.59 | 3.44 | 3.12 | — | — | |
| TokenSESystem Architecture/Backbone=Mamba (Bi)2026.04 | 3.49 | 4.12 | 3.233 | — | — | |
| ANYENHANCE2025.09 | 3.488 | 3.977 | 3.161 | — | — | |
| TokenSESystem Architecture/Backbone=MHSA + Mamba (Bi)2026.04 | 3.483 | 4.11 | 3.219 | — | — | |
| SBMInference steps=12025.10 | 3.459 | 4.106 | 3.198 | 3.742 | 3.937 | |
| cNAC-SEModel Category=VQ-Based Models, Protocol=Fine-Tuned2026.06 | 3.45 | 4.12 | 3.19 | — | — | |
| Discriminative cNAC-SEModel Category=Discriminative2026.06 | 3.45 | 4.09 | 3.18 | — | — | |
| MaskSR-M2025.09 | 3.43 | 4.025 | 3.136 | — | — | |
| MaskSR-MSystem Architecture/Backbone=MaskSR-M [57]2026.04 | 3.43 | 4.025 | 3.136 | — | — | |
| StoRM2025.09 | 3.41 | 3.379 | 2.94 | — | — | |
| StoRMModel Category=Diffusion Models2026.06 | 3.41 | 3.38 | 2.94 | — | — | |
| MaskSR-SSystem Architecture/Backbone=MaskSR-S [57]2026.04 | 3.398 | 4.011 | 3.103 | — | — | |
| FRCRNSystem Architecture/Backbone=FRCRN [56]2026.04 | 3.371 | 3.978 | 3.037 | — | — | |
| dNAC-SEModel Category=VQ-Based Models, Protocol=Fine-Tuned2026.06 | 3.37 | 4.07 | 3.09 | — | — | |
| SB-NCSN++(50)Inference steps=502025.10 | 3.369 | 3.939 | 3.052 | 3.685 | 3.604 | |
| TokenSESystem Architecture/Backbone=Transformer2026.04 | 3.358 | 4.026 | 3.072 | — | — | |
| SB-NCSN++(10)Inference steps=102025.10 | 3.332 | 3.933 | 3.024 | 3.723 | 3.865 | |
| TokenSESystem Architecture/Backbone=Mamba (Uni)2026.04 | 3.329 | 4.018 | 3.049 | — | — | |
| ZipEnhancer2025.10 | 3.323 | 3.918 | 3.002 | 3.639 | 3.264 | |
| FM-MambaInference steps=12025.10 | 3.32 | 4.031 | 3.052 | 3.562 | 3.797 | |
| SGMSEModel Category=Diffusion Models2026.06 | 3.3 | 2.9 | 2.79 | — | — | |
| SB-NCSN++(1)Inference steps=12025.10 | 3.297 | 3.924 | 2.979 | 3.514 | 3.541 | |
| SGMSE2025.09 | 3.297 | 2.894 | 2.793 | — | — | |
| Mamba-baseMode=predictive-mapping2025.10 | 3.284 | 3.974 | 2.993 | 3.683 | 3.644 | |
| SB-UFOGenInference steps=12025.10 | 3.255 | 3.838 | 2.911 | 3.485 | 3.196 | |
| DEMUCSSystem Architecture/Backbone=DEMUCS [15]2026.04 | 3.227 | 3.986 | 2.946 | — | — | |
| SBCTMInference steps=12025.10 | 3.224 | 3.841 | 2.894 | 3.506 | 3.647 | |
| CDiffuSEModel Category=Diffusion Models2026.06 | 3.2 | 3.1 | 2.78 | — | — | |
| SE-CEModel Category=VQ-Based Models2026.06 | 3.2 | 3.81 | 2.86 | — | — | |
| Log-MMSESystem Architecture/Backbone=Statistical baseline2026.04 | 3.153 | 2.869 | 2.424 | — | — | |
| TokenSESystem Architecture/Backbone=Transformer (causal)2026.04 | 3.137 | 3.74 | 2.743 | — | — | |
| Conv-TasNet2025.09 | 3.102 | 2.975 | 2.41 | — | — | |
| Noisy2025.09 | 3.053 | 2.51 | 2.255 | — | — | |
| UnprocessedSystem Architecture/Backbone=None2026.04 | 3.053 | 2.509 | 2.255 | — | — | |
| Noisy2026.06 | 3.05 | 2.51 | 2.26 | — | — |