Speaker Authentication Using Vector Quantization

In this paper, the role of the vector quantization in the speaker authentication system was studied. Vector quantization based speaker authentication system was considered in two phases; training and testing. The training phase concerned with enrolling the speaker models to build the codebook. The codebook generated from a set of feature vectors belong to each sample of speaker s voice. The testing phase includes matching the unknown input speaker with the models. The matching is performed by evaluating the similarity measure between the unknown speech sample and the models in the speaker database to authenticate the input speaker. A weighted similarity measure was introduced; it takes into regard the correlations between the known models in the database. Larger weights are assigned to vectors that have high discriminating power between the speakers and vice versa. The proposed system gave an encourage results; the authentication rate was about 86.6% during a time 4 s.