Automatic music classification and the importance of instrument identification

This paper empirically demonstrates the particular effectiveness of features based on instrumentation in the realm of automatic music classification. The effectiveness of high-level features in general, relative to low-level signal-processing based features, is also demonstrated through an experiment involving automatic genre classification. Experimental evidence is also provided supporting the use of large feature sets combined with feature weighting systems. The tools used to pursue this research are described, as is pertinent background information. An overview is provided of a large library of useful musical features that is available for research purposes to theorists, musicologists and other researchers. The Bodhidharma symbolic music classification system is also introduced as a useful and easy to use tool for researchers wishing to pursue research in music classification based on instrumentation and other high-level features.