A Hybrid Reliable Algorithm for Speaker Recognition Based on Improved DTW and VQ by Genetic Algorithm in Noisy Environment

in this paper a hybrid reliable algorithm is presented which used DTW and VQ optimization plan based on genetic algorithm for text dependent and text independent speaker identification in noisy environment. After preprocessing and feature extraction, to remove the effect of microphone type on saved data, the Cepstral Mean Subtraction (CMS) coefficients have been used. Dynamic Time Warping (DTW) and Vector Quantization (VQ) have been used with Linear Predictive Cepstral Coefficients (LPCC), ÄLPCC, ÄÄLPCC, CMS and Mel Frequency Cepstral Coefficients (MFCC), ÄMFCC, ÄÄMFC, CMS respectively. The Linde-Buzo-Grey (LBG) algorithm has been used to calculate the code book of speakers. Finally genetic algorithm (GA) has been used for optimization and enhancement. The algorithm found optimum coefficients as weights in DTW and calculate best centroids (code books) in VQ algorithm as well. The performance of the algorithm was evaluated by using 25 speakers (15 male and 10 female) for training and testing in noisy and normal condition and experimental results was shown that 0% identification Error Rate was achieved.