Untrained Musical Sound Source Separation
by Unsupervised Density Network

A clarinet plays a single scale:

Clarinet Scale
Source Audio

The clarinet solo is combined with 14 different samples of naturalistic background sounds (like horns honking, phones ringing, traffic passing):

Clarinet & Noise Combined Audio
Input

An untrained, unsupervised Density Network learns, recognizes and extracts the clarinet solo:

Density Network Output
Output