GA-based parameterization and feature selection for automatic music genre recognition
Marcin Serwach, Bartłomiej Stasiak · 2016
Automatic music genre recognition can be done by collaborative filtering or by content-based filtering. In collaborative filtering the music is classified on the basis of the similarity to pieces already classified by users - it is implicitly assumed that the users have proper knowledge to recognize music genres. The second approach - the content-based filtering - is based on extracting sound features directly from music and using them for classification. This study presents a content-based classification system designed to assign music tracks to genres. The classification process is done by 2 classifiers: a feed-forward neural network and k-nearest neighbors algorithm. The structure of the neural network, the number of the nearest neighbors and the actual sound features are selected by a genetic algorithm (GA). The presented approach is tested on a database comprising 10 main genres and 33 subgenres.