Algorithm info: SCNet XL IHF 4 stems (Trained only on MUSDB18)
https://github.com/ZFTurbo/Music-Source-Separation-Training/blob/main/docs/pretrained_models.md#multi-stem-models-musdb18hq
https://github.com/ZFTurbo/Music-Source-Separation-Training/releases/tag/v1.0.15
Metrics:
Metric sdr for vocals: 9.6896
Metric si_sdr for vocals: 8.9874
Metric l1_freq for vocals: 34.3503
Metric log_wmse for vocals: 12.8370
Metric aura_stft for vocals: 8.0442
Metric aura_mrstft for vocals: 9.3154
Metric bleedless for vocals: 22.7374
Metric fullness for vocals: 15.9333
Metric sdr for bass: 11.9436
Metric si_sdr for bass: 10.9242
Metric l1_freq for bass: 52.8340
Metric log_wmse for bass: 16.8948
Metric aura_stft for bass: 4.9780
Metric aura_mrstft for bass: 4.5033
Metric bleedless for bass: 23.0180
Metric fullness for bass: 20.1038
Metric sdr for drums: 11.5839
Metric si_sdr for drums: 10.8247
Metric l1_freq for drums: 37.1489
Metric log_wmse for drums: 15.3002
Metric aura_stft for drums: 8.2920
Metric aura_mrstft for drums: 8.0601
Metric bleedless for drums: 24.4489
Metric fullness for drums: 16.0052
Metric sdr for other: 6.4873
Metric si_sdr for other: 5.4261
Metric l1_freq for other: 32.6762
Metric log_wmse for other: 11.4432
Metric aura_stft for other: 6.2254
Metric aura_mrstft for other: 8.6703
Metric bleedless for other: 19.0048
Metric fullness for other: 16.3093
Date added: 2025-06-18 |
