| The digitization of audio media has brought with it the questions of how to store and catalog it, what to do with the information and how to protect its authorization. Music Information Retrieval is concerned with using retrieval techniques to perform searches on music media. An important problem in music information retrieval is the classification of music. Automatic classification at a coarse level, such as distinguishing among classical, rock and jazz music, is not a difficult problem, but more fine-grained distinctions among musical pieces sharing similar characteristics are more difficult to establish. In the realm of Music Information Retrieval (MIR), there is a burgeoning interest in automatic song and artist identification from audio data. Such work would obviously be useful for anyone who wants to ascertain the performing or composing artists of a new piece of music. It could be also used for automatic music categorization, for a digital library database or the data management for Karaoke according to artists. It could aid preference-based search and recommendations for music. Another area where artist identification may benefit is copyright protection and enforcement. The artist identification problem can be divided into two parts: how to select promising audio features and what kind of machine learning techniques are suitable for the specific task. We implement a prototype system which can analyze features and predict the performing artist of a music piece using machine learning algorithms. We experiment with features and efficient 1-class learning techniques to build customized classifiers for the specific task of artist identification. |
