Design of an Automatic Speaker Recognition System Based on Adapted MFCC and GMM Methods for Arabic Speech

This paper presents a design of an automatic speakerindependent speech recognition system based on adapted Mel Frequency Cepstrum Coefficients (MFCC) associated to Gaussian Mixture Model (GMM) Methods. This experimental study which has performed for various learning times was conducted around MATLAB (R) 7 language environment. Firstly our goal is to design a robust system that is able to identify any Arabic speaker with a good performance in order to implement it later as the embedded system for access control to high secure areas. Results of the experiments using 72 Arabic speakers indicate that recognition error ratio of 2.15 percent or less can be reaches if the learning and the test utterances times are superiors respectively to ten and five seconds.