Title: A music genre classifier combining timbre, rhythm and tempo models
Abstract: The changing music landscape demands new ways of searching, organizing and recommending music to consumers. Content-based music similarity estimation offers a robust solution using a set of audio features. In this paper, we describe the feature extractors to model timbre, rhythm and tempo. We discuss the corresponding feature similarity relations and how the distance measures are combined to quantify music similarity. The proposed system was submitted to 2011 Music Information Retrieval Evaluation eXchange (MIREX) Audio Music Similarity task for validation. Both objective and subjective tests show that the systems achieved an average genre classification of accuracy of 50% across ten genres. Furthermore, the genre classification confusion matrix revealed that the system works best on rap, hiphop and related types of music.