Search

Abstract: Classification in music analysis involves the segmentation of a musicpiece and the categorisation of the segments depending onsimilarity-based criteria. In this paper we investigate, based on aformal approach, how variations in the representation of the musicalsegments and in the categorisation algorithm influence the outcome ofthe classification. More specifically, we vary the choice of featuresdescribing each segment, the way these features are represented, andthe categorisation algorithm. At the same time, we keep the otherparameters, that is the overall model architecture, the music pieces,and the segmentation, fixed. We show that the choice andrepresentation of the features, but not the specific categorisationalgorithm, have a strong impact on the obtained analysis. Weintroduce a distance function to compare the results of algorithmicand human classification, and we show that an appropriate choice offeatures can yield results that are very similar to a humanclassification. These results allow an objective evaluation ofdifferent approaches to music classification in a uniform setting.