I'm guessing a deep learning approach could correct those badly mastered tracks. At least, it is pretty easy to find and generate training data.

(?)