Algorithm info: 1. BS Roformer, unwa, Large v1, weights https://huggingface.co/jarredou/unwa_bs_roformer/tree/main
See also, https://github.com/ZFTurbo/Music-Source-Separation-Training/blob/0a8effedd8e2f403e2a8b3b8f2e9450ae518b4bd/docs/gui.md
2. Mel-Band-Roformer Vocal model, KimberleyJensen
Weights https://github.com/KimberleyJensen/Mel-Band-Roformer-Vocal-Model
stereo sequences --111111
[0, 1, 2, 4, 7, 9, 13, 23]
dos llaves x 0.1
Estimations: gross distinctions
singleTestsPrototypeB
AA=60;[30, 31, 44]
halfwave1024
N14e6
Hunter Hogan
Metrics:
Metric sdr for instrum: 12.7618
Metric si_sdr for instrum: 12.5663
Metric l1_freq for instrum: 45.8790
Metric log_wmse for instrum: 14.8904
Metric aura_stft for instrum: 11.8857
Metric aura_mrstft for instrum: 14.7347
Metric bleedless for instrum: 43.2860
Metric fullness for instrum: 28.5575
Metric sdr for vocals: 13.0580
Metric si_sdr for vocals: 12.6640
Metric l1_freq for vocals: 45.4401
Metric log_wmse for vocals: 14.8904
Metric aura_stft for vocals: 7.6868
Metric aura_mrstft for vocals: 9.1497
Metric bleedless for vocals: 43.4008
Metric fullness for vocals: 18.6787
Date added: 2024-11-18 |
