Music Genre Classification System Based On Music Content

With the rapid expansion of available digital music on the internet, Music InformationRetrieval（MIR）has been developed into a very hot and challenging topic and the automatic musicclassification is one important area of MIR. The task of music genre classification is put the musicinto a particular music style by processing the data signal. The music genre class ification systemcan help users find relevant style music and make more effectively manage the music database.In this paper, the music genre classification algorithm has been successfully implemented andthe basic theory and background about music genre classification are presented. Here focus on themusic feature extraction and SVM classifier algorithm. Based on studying them, this thesis achievesmusic genre classification system and the conducts relative experiments. The main researchcontains the following aspects:(1) After studying the relevant theoretical knowledge on content-based music genreclassification, it gives an overview framework and summarization about the search system.(2) For music feature extraction module, it creatively introduces a robust feature thatPerceptual Linear Prediction (PLP), which is filtered by Log-RASTA (log-Relative Spectral) toremove spectrum rapidly change between frames besides MFCC. Meanwhile get the mean andvariance of the MFCC and RASTA-PLP parameters to represent the music.(3) Having analyzing the statistical theory of SVM and the SVM for dichotomy, it describesthe strategies for multi-classification SVM. The “1-v-r” is applied which has non-classified regions.Here proposes fuzzy membership function to overcome the shortcomings and the experiment resultsshow that this method improves the classification efficiency.(4) According to the feature extraction and SVM algorithm researched，a music genreclassification system is designed and implemented in mat lab and c separately. The system caneffectively classify music genre.