Genre classification of symbolic pieces of music

Abstract

Automatic classification of music is a complex and interesting research problem due to the difficulties that arise when determining the musical features that should be considered for classification and the characteristics that define each particular genre. In this article, we propose an approach for automatic genre classification of symbolic music pieces. We evaluated our approach with a dataset consisting of 225 pieces using a taxonomy of three genres and nine subgenres. Results demonstrate that by only extracting a small set of features from the MIDI files, we are able to obtain better results than competing approaches that use one hundred features for classification.