Performance Evaluation of Statistical Approaches for Automatic Text-Independent Speaker Recognition Using Robust Features

This paper introduces the performance evaluation of statististical approaches for Automatic-text-independent Speaker Recognition system. Automatic-text-independent Speaker Recognition system is to quickly and accurately identify the person from his/her voice. The study on the effect of feature vector size for good speaker recognition demonstrates that the feature vector size in the range of 18-22 can capture speaker related information effectively for a speech signal sampled at 16 kHz. It is demonstrated that the timing varying speaker related information can be effectively captured using Hidden Markov Models (HMMs) than GMM.