Music Emotion Classification Based On Regression

In recent years, the explosive growth of online music resources makes the retrieval and classification of music information become very difficult and has attracted many researchers’ considerable attention. Among this area, automatic music emotion classification(MEC), which is to classify the music data into different categories on the basis of the emotional expression, is consistent with users’ searching habits. However, the ambiguity and subjectivity of emotion makes automatic music classification difficult. In this study, we try finding a new MEC method to improve this music classification performance.In our new method, the Thayer’s emotional model was selected as the basic music emotion database and the music was divided into five categories according to the MIREX standard. Also, the emotion was modeled as two continuous variables, arousal(A) and valence(V) values(AV values)respectively. Then, the MEC was transferred to a regression problem.The MEC system consists of training and testing parts. In the training part, the multiple linear regression, support vector regression(SVR), and k-plane piecewise regression were adopted to evaluate the regression model. Notably, the k-planes piecewise regression was proposed according to k-planes clustering algorithm, which can obtain the hyper plane directly through many times of iteration to avoid calculating the breakpoints firstly in the traditional piecewise regression. In the testing part, after obtaining VA values predicted by the regression model, the music data was classified. The classification accuracy rate was selected to evaluate the system performance. The results revealed that the accuracy rate of the combined using of SVR and k-planes was better than that of using one of the methods alone, which was increased by 3% to 4%. Compared with the MEC based on traditional method(SVM), the regression classification accuracy rate of this method was increased by 6%.