Music genre classification using audio features, different classifiers and feature selection methods


Yaslan Y. , Çataltepe Z.

IEEE 14th Signal Processing and Communications Applications, Antalya, Turkey, 16 - 19 April 2006, pp.535-536 identifier

  • Publication Type: Conference Paper / Full Text
  • City: Antalya
  • Country: Turkey
  • Page Numbers: pp.535-536

Abstract

In this paper, performance of different classifiers (Fisher, linear, quadratik, Naive Bayes, Parzen, k-nearest neighbor) to determine the genre of a given music piece, using different audio feature sets, is examined. For each classifier, performances of feature sets obtained by feature selection and dimensionality reduction methods are also evaluated. Finally, classification accuracy is improved by combining different classifiers. A 10 genre data set of 1000 pieces is used in the experiments. Using a set of different classifiers, a test genre classification accuracy of around 79.6 +/- 4.2 is obtained. This performance is better than 71.1 +/- 7.3% which is the best that has been reported on this data set. Also, by combining different classifiers 80% classification accuracy is obtained.