Contrast of Gaussian Mixture Model and Clustering Algorithm for Singer Identification

Abstract

The intension is to provide the contrast between Clustering Algorithm and Gaussian Mixture Model using Perceptual Linear Prediction features to assess the singer identification structure using two phases, phase 1 as training and phase 2 as testing over the film tracks(vocal with background music). The intent of assessing of singer is to categorize different singers impartial of data that is trained in phase 1. The aspects for two phases are executed for downright tracks from films for 20 different singers. In phase 1 aspects, for individual singer 15 tracks are loaded as input data. Now loaded datas are shaped to go through a deck of pre-handling steps. The pre-handling steps includes three more internal stages with stage1 as Pre-emphasis, stage2 as Frame Blocking and stage3 as Windowing. From individual context of pre-handled signal PLP features are evolved. Using the K-Means Clustering Algorithm and GMM the phase1 output is developed for individual singers. In Clustering algorithm the singer is categorized deployed with choice of the model that gives mean value as minimum. In GMM, by using Maximum Likelihood (ML) algorithm singers are categorized deployed with choice of the model that gives maximum likelihood. Depending on identity of accuracy the singer identification structure is performed. (Abstract)