Performance Evaluation of Voice Classifier Algorithms for Voice Recognition Using Hidden Markov Model

Abstract

This paper provides performance evaluation of K mean and Gaussian mixture  algorithms which are voice classifier algorithms for voice recognition using the differences in their recognition , training and testing time as parameter for the evaluation.

The  performance evaluation results has shown classification efficiency of  K – means &  Gaussian Mixture algorithms. In the results, comparing the Average Training time for Kmeans algorithm: (Standard  database = 435.6854s, Local Database = 411.4578s) while for Gaussian mixture algorithm : (Standard Database = 454.5678s, Local Database = 424.5673s). Moreover, in the considering the Average Testing time, Kmeans algorithm: (Standard database = 23.7178s, Local Database = 23.7178s) while for Gaussian mixture algorithm : (Standard Database = 25.1271s, Local Database = 20.1271s). For the Average Recognition time,  Kmeans algorithm: (Standard  database = 0.3388s, Local Database = 0.3388s) while for Gaussian mixture algorithm : (Standard Database = 0.4345s, Local Database = 0.4345s).

Therefore, conclusions could be made that K-mean algorithm is a better classifier for voices in a voice recognition system because it has minimum training, testing and recognition time compared to Gaussian mixture algorithms.

Key Words: Evaluation, Classification, Efficiency, Algorithm, K- means algorithm, Gaussian mixture algorithm, Training, Speaker, Recognition