https://github.com/deezer/spleeter says the model is "pre-trained" against MUSDB18. I'm trying to get an idea of how much the results will differ / improve by training with the MUSDB18-HQ data. Sometimes I hear artifacts on 2-stem at 11k or 16k on a basic track with say acoustic guitar and vocals. Looking to improve my results. Any thoughts or feedback is greatly appreciated!
https://github.com/deezer/spleeter says the model is "pre-trained" against MUSDB18. I'm trying to get an idea of how much the results will differ / improve by training with the MUSDB18-HQ data. Sometimes I hear artifacts on 2-stem at 11k or 16k on a basic track with say acoustic guitar and vocals. Looking to improve my results. Any thoughts or feedback is greatly appreciated!