Music Information Retrieval of Carnatic Songs Based on Carnatic Music Singer Identification

Abstract

In this paper, a methodology for Carnatic music singer identification is proposed and implemented. The motive behind identifying the singer is to extend this work for efficient music information retrieval of Carnatic music song based on singer identification. In this work, a new set of features that can be used in the process of Carnatic music singer identification is proposed and implemented. The features are based on Carnatic music characteristics like just tempered, varying pitch, varying interval of the octave, etc. The coefficients are called as Carnatic interval cepstral coefficients (CICC) since they are based on Carnatic music's octave interval. The input signal is segmented and frequency of each segment is determined which is followed by determining of the Carnatic interval cepstral coefficients. These coefficients are used to construct a GMM model which forms the basis for singer identification. We have compared our proposed coefficients, CICC with mel-frequency cepstral coefficients for the task of Carnatic singer identification.