In this paper we describe an unsupervised, deterministic algorithm for segmenting DJ-mixed Electronic Dance Music streams (for example; podcasts, radio shows, live events) into their respective tracks. We attempt to reconstruct boundaries as close as possible to what a human domain expert would engender. The goal of DJ-mixing is to render track boundaries effectively invisible from the standpoint of human perception which makes the problem difficult.

We use Dynamic Programming (DP) to optimally segment a cost matrix derived from a similarity matrix. The similarity matrix is based on the cosines of a time series of kernel-transformed Fourier based features designed with this domain in mind. Our method is applied to EDM streams. Its formulation incorporates long-term self similarity as a first class concept combined with DP and it is qualitatively assessed on a large corpus of long streams that have been hand labelled by a domain expert.