Evaluation of n-gram-based classification approaches on classical music corpora

Abstract

The paper deals with evaluation of various n-gram-based composer classification algorithms. Our analysis has a broad scope: We have analyzed three labelled corpora, five similarity measures, several feature extraction methods, the influence of forced balanced training and an extensive range of n-gram lengths. We found that most of the approaches we analyzed, when properly parametrized, can give very good results, on par with other state-of-the art data mining techniques and greatly outperforming humans in composer recognition. © 2013 Springer-Verlag.