ft from https://github.com/ZFTurbo/Music-Source-Separation-Training/releases/download/v1.0.11/model_mel_band_roformer_ep_5_sdr_8.9443

That Metrics value is here Instr drums sdr: 10.4314 (Std: 3.8924) Instr bass sdr: 7.7418 (Std: 4.3232) Instr other sdr: 7.0982 (Std: 2.9073) Instr vocals sdr: 10.6979 (Std: 3.3886) Metric avg sdr : 8.9923 from musdb18 test dataset


I uploaded ver.2 today (about 3/2 pm2 in Japan) Maybe, SDR was to lower, but quality was uped at all.... just feel it. This is model for extract 4stem. Do not use for extract instruments.

Instr drums sdr: 10.2869 (Std: 3.8291) Instr drums fullness: 17.1690 (Std: 7.1304) Instr drums bleedless: 32.6756 (Std: 9.4637) Instr bass sdr: 7.6155 (Std: 4.3410) Instr bass fullness: 11.6789 (Std: 7.6093) Instr bass bleedless: 38.1782 (Std: 20.6632) Instr other sdr: 7.0061 (Std: 2.7807) Instr other fullness: 16.9378 (Std: 6.5542) Instr other bleedless: 29.2958 (Std: 12.1943) Instr vocals sdr: 10.5878 (Std: 3.3455) Instr vocals fullness: 19.6140 (Std: 7.8615) Instr vocals bleedless: 33.0137 (Std: 17.0555) Metric avg sdr : 8.8741 Metric avg fullness : 16.3499 Metric avg bleedless : 33.2908

Downloads last month
136
Inference Providers NEW
This model is not currently available via any of the supported Inference Providers.
The model cannot be deployed to the HF Inference API: The model has no library tag.

Space using SYH99999/MelBandRoformer4StemFTLarge 1