A Study On Music Emotion Classification

With the development of the Internet and the change of storage methods in recent years, peopleget music resources from Internet become easier. But it is more difficult to get the music which theyare interested in. How to manage massive music effectively is imminent. Therefore, this study ofmusic emotion classification has great practical significance.Based on the analysis of the emotional model, the music is divided into four categories such ashappy, sad, calm and excited. This paper proposes a double-layers structure of music emotionclassification system. The music is divided into fast-paced music and slow-paced music on the firstlayer. Obviously, happy songs and excited songs belong to fast-paced class, while sad music andcalm songs belong to slow-paced class. If the test song belongs to fast-paced song, epmodel2classifier divides it into happy song and excited song; otherwise the csmodel2classifier divided thesong into sad song and calm song on the second layer.The music classification based on emotion includes feature extraction part, classification modeltraining and testing part. During the feature extraction part, we found the range of feature valuevaried wildly from female singers to male singers. Therefore, this paper designed a experimentcomparing the classify accuracy of music with different gender singer. In the training sessions, Thepaper uses the Adaboost algorithm and SVM algorithm to train different emotional classificationmodel. The results show that Adaboost algorithm is superior to SVM algorithm.