Study On Methods Of Music Style Classification Based On Data Mining Techniques

The fast developing digital information technology facilitates the expansion of musical industry on Internet, music playing and downloading have also been the main service for many enterprises of Internet service. Such facilitation of Internet results in the emergence of large numbers of new musical genres, while users’preference of different styles varies with individual. How to solve the problem of genre classification for music has been a hot issue widely studied by companies and scholars in this era.In this thesis, a method of genre classification for MIDI music file based on musical theory has been proposed by using data mining techniques. For this method, information about harmony and melody originally contained in the music has been regarded as the basis of genre classification respectively. Data mining techniques are used for the bottom-up sectional hierarchical genre classification for music. We got ideal experiment results about the comparisons between the methods proposed in this thesis and other methods for musical genre classification.Some innovations in this thesis include:(1) an melody extraction algorithm for midi file based on threshold and Shannon entropy which is an improvement of Skyline melody extraction algorithm has been proposed in this thesis;(2) algorithms for harmony information extraction has been proposed in this work;(3) this is also the first use of Rough Set Theory in music genre classification area;(4) an algorithm for attribute reduction with complexity of o(|c||U/C|) has been proposed:(5) an initiative searching strategy has been proposed in this thesis for cutting down the overhead of I/O operation and large complexity of Apriori algorithm in the process of frequent pattern mining:(6) for eviting the disturbance from key shifting to the total result of classification, an algorithm of melody segmentation has also been proposed in this work. This work also adopts the thinking of bottom-up for decision making and considers the harmony and melody as different bases respectively in the process of classification.All experiments in this work have been conducted on MATLAB software, and Midi Toolbox developed by Torvaldis has been used for deciphering the MIDI files.