Algorithm info: chunk_size = 485100 (aka dim_t = 1101)
overlap = 2
batch_size= 2
Available here: https://github.com/ZFTurbo/Music-Source-Separation-Training/blob/main/docs/pretrained_models.md
-jarredou
Metrics:
Metric sdr for vocals: 9.1941
Metric si_sdr for vocals: 8.4346
Metric l1_freq for vocals: 34.8315
Metric log_wmse for vocals: 12.4322
Metric aura_stft for vocals: 7.1191
Metric aura_mrstft for vocals: 8.0315
Metric bleedless for vocals: 33.1729
Metric fullness for vocals: 13.4323
Metric sdr for bass: 11.0813
Metric si_sdr for bass: 10.2839
Metric l1_freq for bass: 52.3254
Metric log_wmse for bass: 16.2626
Metric aura_stft for bass: 5.3365
Metric aura_mrstft for bass: 4.0825
Metric bleedless for bass: 34.6858
Metric fullness for bass: 16.0897
Metric sdr for drums: 11.2916
Metric si_sdr for drums: 10.5677
Metric l1_freq for drums: 37.7272
Metric log_wmse for drums: 15.4345
Metric aura_stft for drums: 6.6604
Metric aura_mrstft for drums: 5.9330
Metric bleedless for drums: 37.2818
Metric fullness for drums: 12.4972
Metric sdr for other: 5.9636
Metric si_sdr for other: 4.8734
Metric l1_freq for other: 32.4892
Metric log_wmse for other: 11.0208
Metric aura_stft for other: 4.7026
Metric aura_mrstft for other: 6.7046
Metric bleedless for other: 22.4634
Metric fullness for other: 13.8835
Date added: 2025-01-21 |
