Singer Identification – Analysis with SVM and GMM Classifier

Abstract

Singer Identification (SID) plays a vital role in the music information retrieval (MIR) system, as music and singing are inter-bounded entities and partial without one another. This paper presents the singer identification system that identifies the singer by extracting the acoustic features that completely describe the vocal characteristics of the singing voice using a self-developed cappella database. The paper first discusses the performance of the individual acoustic features and then signifies the impact of its combination on the SID accuracy. The SID was investigated with two classifiers – Support Vector Machine (SVM) and Gaussian Mixture Model (GMM). It was found that for all the combinations of the acoustic features, SVM outperformed GMM. Moreover, the experimental work also revealed that the rbf kernel performed better than the polynomial kernel both in terms of performance and computation cost.