A Robust Music Composer Identification System Based on Cepstral Feature and Models

Abstract

Music is an integral part of everyone’s life. The task of recognizing a composer by listening to a musical piece is difficult for people with no knowledge of music. Even such a task is difficult for people in musical theory because every day a new music composer is born. To address this problem of classification of a musical tune for a large group of music composers, we construct an automatic system that can distinguish music composers based on their tunes composed. Classification of music based on its composer is very essential for faster retrieval of music files. In this paper, we make a large database that consists of several musical tunes belonging to several composers. We extract MFCC features and train the system using the clustering technique by developing a classification model that can accurately classify a musical tune that belongs to a music composer whose tunes are trained. The overall accuracy of the proposed system is 72.5% without discrete wavelet transform (DWT) and 77.5% with DWT.