So I did train the Metokur model on pitch extraction and shit like that for music... I wonder how well that works.