Speech Enhancement on DNS (test)
19.7SI-SDR (dB)Sudo rm-rf
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Sudo rm-rfTraining Mode=Supervised in-domain, U-ConvBlocks=32, Model Params=0.73M, Available Training Data=Clean DNS 100%, Noise DNS 100%, Mixture DNS 100%2022.02 | 19.7 | 2.95 | 0.971 | |
| Sudo rm-rfTraining Mode=Supervised in-domain, U-ConvBlocks=8, Model Params=0.56M, Available Training Data=Clean DNS 100%, Noise DNS 100%, Mixture DNS 100%2022.02 | 18.6 | 2.69 | 0.962 | |
| Supervised StudentSetting=Supervised In-domain, Model=Student (U=8), Params (10^6)=0.56, Clean Speech Ds (DNS)=100%, Clean Noise Dn (DNS)=100%2021.10 | 18.6 | 2.69 | 0.962 | |
| RemixITTraining Mode=Semi-supervised, Role=Student, U-ConvBlocks=8 -> 16 -> 32, Model Params=0.73M, Available Training Data=Mixture DNS 100%2022.02 | 18 | 2.6 | 0.959 | |
| RemixITSetting=Semi-supervised, Model=Student (U=32), Params (10^6)=0.73, Mixture Dm (DNS)=100%2021.10 | 18 | 2.6 | 0.959 | |
| RemixITTraining Mode=Semi-supervised, Role=Teacher, U-ConvBlocks=8, Model Params=0.56M, Available Training Data=Clean Speech LFSD 100%, Noise LFSD 100%2022.02 | 17.6 | 2.61 | 0.958 | |
| RemixITTraining Mode=Semi-supervised, Role=Student, U-ConvBlocks=8, Model Params=0.56M, Available Training Data=Mixture DNS 100%2022.02 | 17.6 | 2.52 | 0.956 | |
| RemixITSetting=Semi-supervised, Model=Teacher (U=8), Params (10^6)=0.56, Clean Speech Ds (DNS)=100%, Clean Noise Dn (LFSD)=100%2021.10 | 17.6 | 2.61 | 0.958 | |
| FullSubNet*Training Mode=Supervised in-domain, Model Params=5.6M, Available Training Data=Clean DNS 100%, Noise DNS 100%, Mixture DNS 100%2022.02 | 17.3 | 2.78 | 0.961 | |
| FullSubNetSetting=Supervised In-domain, Params (10^6)=5.6, Clean Speech Ds (DNS)=100%, Clean Noise Dn (DNS)=100%2021.10 | 17.3 | 2.78 | 0.961 | |
| RemixITTraining Mode=Unsupervised, Role=Student, U-ConvBlocks=8 -> 16 -> 32, Model Params=0.73M, Available Training Data=Mixture DNS 100%2022.02 | 16 | 2.34 | 0.952 | |
| RemixITSetting=Unsupervised, Model=Student (U=32), Params (10^6)=0.97, Mixture Dm (DNS)=100%2021.10 | 16 | 2.34 | 0.952 | |
| RemixITTraining Mode=Unsupervised, Role=Student, U-ConvBlocks=8, Model Params=0.56M, Available Training Data=Mixture DNS 100%2022.02 | 15.5 | 2.27 | 0.947 | |
| RemixITTraining Mode=Unsupervised, Role=Teacher, U-ConvBlocks=8, Model Params=0.79M, Available Training Data=Noise LFSD 20%, Mixture DNS 80%2022.02 | 14.8 | 2.15 | 0.94 | |
| RemixITSetting=Unsupervised, Model=Teacher (U=8), Params (10^6)=0.79, Clean Noise Dn (LFSD)=20%, Mixture Dm (DNS)=80%2021.10 | 14.8 | 2.15 | 0.94 | |
| MixITBackbone=Sudo rm-rf, U-ConvBlocks=8, Model Params=0.79M, Noise Config=Extra OOD noise, Available Training Data=Noise LFSD 50%, Mixture DNS 100%2022.02 | 14.5 | 2.03 | 0.93 | |
| MixITSetting=Unsupervised, Variant=Extra noise [9], Model=Student (U=8), Params (10^6)=0.79, Clean Noise Dn (LFSD)=50%, Mixture Dm (DNS)=100%2021.10 | 14.5 | 2.03 | 0.93 | |
| MixITBackbone=Sudo rm-rf, U-ConvBlocks=8, Model Params=0.79M, Noise Config=In-domain noise, Available Training Data=Noise DNS 20%, Mixture DNS 80%2022.02 | 14.4 | 2.13 | 0.933 | |
| MixITSetting=Unsupervised, Variant=In-domain, Model=Student (U=8), Params (10^6)=0.79, Clean Noise Dn (DNS)=20%, Mixture Dm (DNS)=80%2021.10 | 14.4 | 2.13 | 0.933 | |
| MixITBackbone=Sudo rm-rf, U-ConvBlocks=8, Model Params=0.79M, Noise Config=OOD noise, Available Training Data=Noise LFSD 20%, Mixture DNS 100%2022.02 | 14.3 | 2.02 | 0.933 | |
| MixITSetting=Unsupervised, Variant=OOD noise, Model=Student (U=8), Params (10^6)=0.79, Clean Noise Dn (LFSD)=20%, Mixture Dm (DNS)=100%2021.10 | 14.3 | 2.02 | 0.933 | |
| Input Noisy MixtureDescription=Raw noisy audio input2022.02 | 9.2 | 1.58 | 0.915 | |
| Input Noisy Mixture2021.10 | 9.2 | 1.58 | 0.915 |