Music-Source-Separation-Training
Music-Source-Separation-Training copied to clipboard
Differences between mdx23c model and Kuielab model results
Hello
I was wondering if anyone knows what the differences are between the mdx23c model in this repo and the one from Kuielab’s sdx23 repo (https://github.com/kuielab/sdx23). As far as I can tell, they’re supposed to be the same model, but when I train them on MUSDB18-HQ, I get much lower results with the one from this repo compared to the Kuielab version.
Thanks!
Models are the same (code from original repo used almost as is), but training parameters and config files can be different.
Sounds good thank you!