Optimalization of GMM for Text Independent Speaker Verification System

This paper presents automatic verification system based on the voice recognition technology. System works as text independent and is designed to verify a person using a short utterance. GMM method is used for speaker modeling and GMM-UBM classifier is used for process of decision. The main goal of experiments in this paper is to finding the optimum size of the GMM and UBM models (optimum numbers of components). Experimental evaluation of the system is conducted on the database of 24 people, where around 6 minutes of speech data from one person is available for training and 3 types of utterances with different lengths from 1 to 9 seconds are available for testing.