Music mood classification using audio power and audio harmonicity based on MPEG-7 audio features and Support Vector Machine

Abstract

Music can affect a person's mood. Music psychologists agree that music has a significant impact on a person's mood that determines their behavior. Therefore, our research examines the audio features that affect mood. Our method is to perform feature extraction based on MPEG-7 Low-Level Descriptors. MPEG-7 is international standardized multimedia metadata in ISO/IEC 15938. In this paper, we have made a researched about music mood classification using Audio Power and Audio Harmonicity features. The result of the extraction of the MPEG-7 obtained 17 features low-level descriptors. These features are classified using Support Vector Machine (SVM). There are two stages of SVM: training and prediction phase. Traning phase is when the machine learns to recognize the characteristics of the signal on a label while in prediction phase, it gives the predicted outcome of a label on a new signal characteristic pattern. The success rate of this experiment was 74.28% using Audio Power and Audio Harmonicity, 37.14% using Audio Spectrum Projection, and 28.57% using Audio Power, Audio Harmonicity and Audio Spectrum Projection.