Automatic Music Mood Detection through Musical Structure Analysis

Abstract

Since music has a close relation with people emotion, many researchers have studied on detecting music mood. Usually, they manually selected the representative segment of music and classified its mood. Although such manual approaches correctly detect the mood, their manual intervention makes their application for new music difficult. Moreover, the mood variation in music makes their application more difficult. To cope with these problems, we present an automatic method to classify the music mood. After the music is segmented into groups having similar characteristics by structural information, the support vector regressors are constructed with musical features from pitch, beat, and spectrum of each segment. The mood of unknown music is predicted by the regressors. Experimental results show that performance of the proposed approach is comparable to the performances of manual approaches.