An empirical evaluation of computational and perceptual multi-label genre classification on music.

摘要

Music genre classification is a high-level task in Music Information Retrieval (MIR). It refers to the process of assigning genre labels to music for various tasks, including, but not limited to categorization, organization and browsing. Due to the subjective and ambiguous nature of music, previous single-label classification is inadequate.;In this thesis, we study multi-label music genre classification from perceptual and computational perspectives. First, we design a set of perceptual experiments to investigate genre-labelling behavior by individuals. The results from these experiments lead us to speculate that multi-label classification is more appropriate for classifying music genres. Second, we then design a set of computational experiments to evaluate multi-label classification algorithms on music data. These experiments not only support our speculation but also reveal which algorithms are suitable for music genre classification. Finally, we propose a group of ensemble approaches for combining multi-label classifiers to further improve classification performance. Our approaches are evaluated for their effectiveness.