Segmentation and Timbre Similarity in Electronic Dance Music

In this paper we argue that the notion of music similarity should be expanded into sub-similarities, meaning that similarity of music has to be judged with respect to a certain context, such as melody, harmony, rhythm or timbre. We start by focusing on timbre similarity, restricted to the domain of Electronic Dance Music (EDM). We will assess the similarity of segments of music, thus we start by studying segmentation before we come to the topic of similarity. The segmentation algorithm performs well on an EDM dataset as well as on a standard MIREX dataset. Initial listening tests of the similarity model give promising results but will have to be further evaluated in future research.