UC San Diego

Machine learning, signal processing and data mining are being combined to analyze audio content in a relatively new field of research called computer audition. This thesis develops and describes a number of computer audition methods and shows how they can be applied to solve challenges including automatic tagging, similarity and recommendation, search and discovery, and segmentation of music content. To achieve these advances in music understanding requires human guidance. A further contribution of this work is to pioneer game-powered machine learning that uses crowdsourced human intelligence to guide the training of machine algorithms. By leveraging human perception with machine automation, the work described in this thesis presents a comprehensive approach to computer audition that leads to the development of machines that understand music