Audio Source Separation
in a Musical Context
The Problem:
Extracting a melody or any part of a song can be daunting and often deemed “impossible”. Having the individual source material would make it easier, but if you had the source, you wouldn’t need the extraction and you almost never have the source anyway. However use of Probabilisitic Latent Component Analysis with a mimic of the target audio can provide substantial results. Take a look.