An OMPGW Method for Feature Extraction in Automatic Music Mood Classification Using PSO-SVM

Abstract

In music related applications, Music mood are very useful. Music signal’s inherent emotional expression are represented music mood. Analysis of Music signal based on emotions are proposed in this paper. Wigner distribution function, Gabor functions and orthogonal matching pursuit forms base of this method. This method is termed as OMPGW and it has three schemes. They are high, middle and low level scheme. Music signal’s adaptive time-frequency decomposition is provided by combining Gabor function with orthogonal matching pursuit in middle low-level schemes.

High temporal and spatial resolutions are given by proposed algorithm when compared to other algorithms. It also provides better representation of structure of music signal. From low-level schemes, results time-frequency energy distribution is obtained by applying Wigner distribution function in middle level schemes. Music mood classification procedure and audio feature modelling are described by high level schemes. Features are modelled using support vector machines with Particle Swarm Optimization classifier. Based on emotion model, features are extracted by proposed method.

Particle swarm optimization is proposed to search optimum parameters. It enhances the support vector machine’s ability in generalization and learning. Four datasets are used for conducting experimentation and proposed method produced a better results. Various mood clusters are formed by classifying music clips in music mood classification with mean accuracy of 69.53%.