Table 1 SDR, SIR, and SAR value of U-Net-based models using different backbone networks, evaluated on the test set of Musdb18
From: Stripe-Transformer: deep stripe feature learning for music source separation
| Â | SDR | SIR | SAR | Â | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Â | Bass | Drums | Other | Vocals | Bass | Drums | Other | Vocals | Bass | Drums | Other | Vocals | Mean |
ResCNNs | 4.99 | 6.41 | 5.23 | 6.74 | 9.65 | 13.19 | 7.77 | 15.51 | 5.88 | 6.22 | 4.63 | 6.73 | 7.75 |
SR-T | 4.81 | 6.57 | 5.09 | 6.85 | 10.28 | 12.73 | 8.09 | 14.62 | 5.92 | 6.47 | 4.55 | 6.98 | 7.75 |
Stripe-T | 5.46 | 7.56 | 5.82 | 7.75 | 11.00 | 13.36 | 9.00 | 16.37 | 5.92 | 7.23 | 5.18 | 7.57 | 8.52 |