Multi-instrument Source Separation on MUSDB18 SiSEC (test)
3.46Vocals SDRWavenet-based model
Evaluation results
| Method | Links | ||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Wavenet-based modelstacks=32018.10 | 3.46 | 11.26 | 5.18 | 4.39 | 13.37 | 5.08 | 2.24 | 6.36 | 5.94 | 0.54 | 4.07 | 4.41 | |
| Wavenet-based modelstacks=42018.10 | 3.35 | 11.25 | 5.24 | 4.13 | 13.23 | 5 | 2.49 | 6.53 | 5.77 | 0.41 | 3.83 | 4.47 | |
| Wavenet-based modelstacks=4, dissimilarity loss (Ld)=true2018.10 | 3.05 | 10.58 | 4.8 | 4.09 | 12.85 | 5.31 | 2.23 | 5.66 | 6.37 | -0.19 | 4.37 | 3.24 | |
| Wavenet-based modelstacks=52018.10 | 2.84 | 9.56 | 5.2 | 4.6 | 12.66 | 6.08 | 2.48 | 6.7 | 6.27 | 0.18 | 3.26 | 4.75 | |
| DeepConvSepsampling rate=16kHz2018.10 | 2.38 | 4.45 | 8.39 | 3.19 | 6.69 | 6.58 | 0.2 | 1.92 | 7.46 | -2.02 | 1.74 | 2.5 | |
| DeepConvSepsampling rate=44kHz2018.10 | 2.37 | 4.65 | 8.04 | 3.14 | 6.73 | 6.55 | 0.17 | 1.98 | 7.06 | -2.13 | 1.84 | 2.33 | |
| Wavenet-based modelstacks=12018.10 | 0.35 | 3.94 | 4.38 | 1.24 | 7.98 | 3.56 | 0.35 | 4.54 | 4.7 | -2.7 | -1.37 | 6.75 | |
| Wavenet-based modelstacks=22018.10 | 0.07 | 4.48 | 3.49 | -0.09 | 6.87 | 2.88 | -0.55 | 0.87 | 7.8 | -2.05 | -0.97 | 8.96 |