10th International Society for Music Information Retrieval Conference (ISMIR 2009) . SMERS: MUSIC EMOTION RECOGNITION USING SUPPORT VECTOR REGRESSION Byeong-jun Han, Seungmin Rho Roger B. Dannenberg Eenjun Hwang School of Electrical Engineering Korea University {hbj1147, smrho}@korea.ac.kr School of Computer Science Carnegie Mellon University rbd@cs.cmu.edu School of Electrical Engr. Korea University ehwang04@korea.ac.kr ABSTRACT Music emotion plays an important role in music retrieval, mood detection and other music-related applications. Many issues for music emotion recognition have been addressed by different disciplines such as physiology, psychology, cognitive science and musicology. We present a support vector regression (SVR) based music emotion recognition system. The recognition process consists of three steps: (i) seven distinct features are ex- tracted from music; (ii) those features are mapped into eleven emotion categories on Thayer’s two-dimensional emotion model; (iii) two regression functions are trained using SVR and then arousal and valence values are pre- dicted. We have tested our SVR-based emotion classifier in both Cartesian and polar coordinate system empirically. The result indicates the SVR classifier in the polar repre- sentation produces satisfactory result which reaches 94.55% accuracy superior to the SVR (in Cartesian) and other machine learning classification algorithms such as SVM and GMM. 1. INTRODUCTION With the recent advances in the field of music informa- tion retrieval, there is an emerging interest in (automati- cally) analyzing and understanding the emotional content of music. Due to the diversity and richness of music con- tent, many researchers have been pursuing a multitude of research topics in this field, ranging from computer science, digital signal processing, mathematics, and sta- tistics applied to musicology and psychology. Many computer scientists [1][2] have focused on music retriev- al by using musical meta-data (such as title, genre or mood) as well as low-level feature analysis (such as pitch, tempo or rhythm), while music psychologists [3][4] have been interested in studying how music communicates emotion. Currently, there is no standard method to measure and analyze emotion in music. However, a psychological model of emotion has found increasing use in computa- tional studies. Thayer’s two-dimensional emotion mod- el [5] offers a simple but quite effective model for plac- ing emotion in a two-dimensional space. In the model, the amount of arousal and valence is measured along the vertical and horizontal axis, respectively The goal of this paper is to develop a music emotion recognition system for predicting the arousal and valence of a song based on audio content. First, we analyzed sev- en different musical features (such as pitch, tempo, loud- ness, tonality, key, rhythm and harmonics) and mapped them into eleven categories of emotion: angry, bored, calm, excited, happy, nervous, peaceful, pleased, relaxed, sad and sleepy. This categorization is based on Juslin’s theory [3] along with Thayer’s emotion model [5]. Se- condly, we adopt support vector regression (SVR) [6] as a classifier to train two regression functions for predict- ing arousal and valence values based on the low-level features, such as pitch, rhythm and tempo, extracted from music. In addition, we compared our SVR-based method with other classification algorithms such as GMM (Gaus- sian Mixture Model) and SVM (Support Vector Ma- chine) to evaluate the performance. In the following section, we present a brief overview on the current state-of-the-art music recognition systems, and emotion models. In Section 3, we illustrate a musical feature extraction scheme and give an overview of our proposed system. Section 4 describes our proposed SVR- based music emotion recognition method. Experimental results are given in Section 5. In the last section, we con- clude the paper with some observations and future work. 2. RELATED WORK Many researchers have explored models of emotions and factors that give rise to the perception of emotion in mu- sic. Many other researchers investigate the problem of automatically recognizing emotion in music. 2.1 Music and Emotion Traditional mood and emotion research in music has fo- cused on finding psychological and physiological factors that influence emotion recognition and classification. During the 1980s, several emotion models were pro- posed, which were largely based on the dimensional ap- proach for emotion rating. Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. © 2009 International Society for Music Information Retrieval 651