Inference using pretrained weights

[igoralvarezz]

Hi @morgan76, I've just finished reading the original paper and I really liked it, congrats on the amazing work!

I'm not an academic on the topic but only a musician with a technical background trying to keep up with the pace of advancements on the fields of DL and ML applied to the MIR and Music Generation tasks. I've found this git repository linked on the original paper.
I was able to create the conda environment, following the instructions of the readme, and install all the dependencies of the project correctly. I also took a brief look on the functions available on the .py files on the repository but couldn't find any function aimed at just inferencing (segment any given song) using the pre-trained weights.

I'd like to check with you if there is a pre-build function for that end and, if not, how would I proceed to implement one (which methods should I use, etc). Would I need to train a model from scratch to do so?

Thanks in advance