Designing of a gender based classifier for western music

Abstract

A musical piece constitutes of vocals and background music which is repetitive in nature. Their separation is an essential job in many applications, like music information retrieval (MIR), gender recognition and lyrical recognition. In this paper, we propose a classifier which cleaves the music/song on the basis of gender of the singer without having to listen to it. The basic idea is the music/vocal using REPET algorithm and using the vocals extracted for parameter extraction. The parameters used here are pitch, ZCR and MFCC. Based on these values, a classifier was designed using fuzzy inference system (FIS). A dataset of 43 songs was prepared to check the validity of the system. These songs were experimented upon and showed the accuracy of 82.6% by using the classifier designed.