Development of a rhythm similarity model for electronic dance music

This research investigates rhythmic similarity in Electronic Dance Music (EDM). Rhythm in EDM is built on the concept of a 'loop', a repeating pattern often associated with a four-measure percussive pattern. In our model, music tracks are segmented using a structural segmentation algorithm developed in previous research, and features are extracted from the audio. A segment is split into different streams of rhythmic patterns, where each pattern is described with three attributes: a) characterization of attacks, b) periodicity of rhythmic elements, and c) metrical distribution of onsets. These attributes for all patterns within a segment contribute to the rhythm characterization of that segment. Finally, an algorithm is developed to measure similarity between segments.