Text-Independent Speaker Identification Using Hybrid Vector Quantization / Gaussian Mixture Models Pattern Classifier

Speaker recognition is a process where a person is recognized on the basis of his/her voice signals. In this paper we provide a brief overview for evolution of pattern classification technique used in speaker recognition. The most common approach to speaker recognition is the use of global Gaussian Mixture Model (GMM). The dominant advantages of GMM approach is that speaker identification can be performed in a completely text independent environment. Besides, GMM are base on probabilistic framework, it provide high-accuracy recognition. However, GMM techniques does not work well in some situation due to it behavior of ignores knowledge of the underlying phonetic content of the speech. To overcome those shortages, the new classification Model is generated. We introduce a new hybrid Vector Quantization / Gaussian Mixture Models (VQ/GMM) model to improve recognition rate of the speaker identification system in the paper. Besides, we also concerns about a comparison performance of hybrid VQ/GMM, DTW, VQ, GMM and SVM techniques for speaker identification. Topics of how we construct hybrid VQ/GMM for speaker identification and experimental result for these 5 techniques are presented in this paper. Experiments in this study were performed using TIMIT speech database. Experimental result shows that hybrid VQ/GMM gain the best result among 5 types of classifier.