Multiple Viewpoint Systems for Music Classification

Abstract

This paper describes a new statistical modelling method for music classification. The method is an extension of the multiple viewpoint method for music prediction and generation. A multiple viewpoint system significantly outperforms all component viewpoints on the tasks of folk tune genre and region classification. The method is successfully applied to predict the genres of unlabelled Basque folk tunes.

Acknowledgments

The Fundación Euskomedia and the Fundación Eresbil are graciously thanked for making the Cancionero Vasco available for study. Thanks to Ruben Hillewaere for compiling the europa-6 and dance-9 datasets, and to Kerstin Neubarth for comments on the manuscript. This research was partially supported by a grant Análisis Computacional de la Música Folclórica Vasca (2011–2012) from the Diputación Foral de Gipuzkoa, Spain.