Algorithm info: 1. BS Roformer (vocals, instrumental) ver 2024.04 (SDR vocals: 11.24, SDR instrum: 17.55)
2. BS Roformer (vocals, instrumental) ver 2024.08 (SDR vocals: 11.31, SDR instrum: 17.62)
3. MelBand Roformer (vocals, instrumental) Bas Curtiz edition (SDR vocals: 11.18, SDR instrum: 17.49)
4. Weights https://github.com/KimberleyJensen/Mel-Band-Roformer-Vocal-Model
stereo sequences --111111
[0, 1, 2, 4, 7, 9, 13, 23]
N32k14 halfwave
A few options were slightly limited because Google Colab kept crashing due to running out of system memory.
Hunter Hogan
Metrics:
Metric sdr for instrum: 17.8213
Metric si_sdr for instrum: 17.7277
Metric l1_freq for instrum: 40.3096
Metric log_wmse for instrum: 14.5455
Metric aura_stft for instrum: 12.2565
Metric aura_mrstft for instrum: 15.7980
Metric sdr for vocals: 11.5139
Metric si_sdr for vocals: 11.1141
Metric l1_freq for vocals: 39.8344
Metric log_wmse for vocals: 14.5455
Metric aura_stft for vocals: 8.7800
Metric aura_mrstft for vocals: 10.7067
Date added: 2024-10-29 |
