Supervised learning for genre classification of audio tracks

Abstract

[en] Music is a form of art that accompanies all of us every day and, with the appearance of on line services such as Spotify or Tidal, music analysis has become crucial to these services to recommend new music to users and to classify all new tracks uploaded every day. In this dissertation, we provide with a system for multi-class classification of genres for audio tracks. We base on standard MFCC audio descriptors to then define a compact audio track feature vector representation. Different machine learning classifiers are tested to perform final genre classification, also providing an analysis of the relevance of the different features.