An approach for identifying salient repetition in multidimensional representations of polyphonic music

SIATEC is an algorithm for discovering patterns in multidimensional datasets (Meredith et al., 2002). This algorithm has been shown to be particularly useful for analysing musical works. However, in raw form, the results generated by SIATEC are large and difficult to interpret. We propose an approach, based on the generation of set-covers, which aims to identify particularly salient patterns that may be of musicological interest. Our method is capable of identifying principal musical themes in Bach Two-Part Inventions, and is able to offer a human analyst interesting insight into the structure of a musical work.