Classification of Musical Playing Styles using MIDI Information

The automation of playing style classification is a crucial step in the analysis of musical information. In this work, “style” is taken to be a more functional description of a specific instrument within a song rather than a description of the genre or articulation in which that instrument is playing. It is shown that by using a supervised learning approach, we can classify the playing style of an instrument as bass, rhythm, lead, or fingerpicking with a very high accuracy. Several machine learning algorithms are tested, the most successful one was the support vector machine with a Gaussian kernel, was found to provide an accuracy of 0.982 using a leave-one-out-crossvalidation technique. Additionally, we use a recursive feature search to analyze which subset of a larger feature set is most important for this classification task.