Singer Identification Using Gaussian Mixture Model (GMM)

Abstract

In this paper, identifying the artist of a query song from the audio database is considered. To build the model of a specific signer, only the vocal segments of a song is employed. Mel-Frequency Cepstral Coefficient (MFCC) is used for extracting salient features of each artist. Classification among a group of artists is performed by Gaussian Mixture Model (GMM classifier. The desired “singer” may be defined as an individual or group who records or performs popular songs under a particular identification name. After extracting the vocal segments, they are fed into the singer identification system that has been trained on data taken from songs of other album by the same artist. Experiments are carried out a group of singers where the songs are in three different genres.