Implement an external who can separate live audio into two streams: harmonic and percussive. In this way, we would be able to transcribe audio from polyphonic music.
Retrain the KNN with percussive sounds separated from commercial recordings or live polyphonic audio.