Algorithm info: this is probably the dumbest idea i’ve ever had for an instrumental model. don’t use it as your main one, some vocals will leak because i added tracks with vocal chops to the dataset. just use this model for songs that have vocal chops
@neoculture
Metrics:
Metric sdr for instrum: 14.3509
Metric si_sdr for instrum: 14.2108
Metric l1_freq for instrum: 35.3563
Metric log_wmse for instrum: 11.4281
Metric aura_stft for instrum: 15.5114
Metric aura_mrstft for instrum: 20.5945
Metric bleedless for instrum: 32.5644
Metric fullness for instrum: 39.8806
Metric sdr for vocals: 8.0434
Metric si_sdr for vocals: 7.3192
Metric l1_freq for vocals: 30.2596
Metric log_wmse for vocals: 11.4281
Metric aura_stft for vocals: 6.5227
Metric aura_mrstft for vocals: 7.2852
Metric bleedless for vocals: 20.8052
Metric fullness for vocals: 13.4828
Date added: 2025-10-07 |
