Text independent spekaer identification system using VQ and GMM

The goal of automatic speaker recognition systems is to extract, characterize and recognize the information in the speech signal conveying speaker identity. In this paper two different modeling techniques Vector Quantization (VQ) and Gaussian Mixture Model (GMM) and their combinations are used for text-independent speaker recognition. Experimental results show that hybrid VQ/GMM method reduces computational time by a significant amount Effects of increasing population, number of mixture used, size of code book, computation time, and effect of duration of training and testing session have been studied and it is observed that it increases the accuracy and robustness of the system when complementary spectral features along with the main features are used with combination of models.