The Speaker Recognition Systmen Based On GMM-UBM Moedl

Speaker recognition forms characteristic parameters of speaker by extract the features of voice and build the acoustic model to recognize the speaker’s identity automatically. Recent decades, science and technology developed rapidly, so peoples life become more and more convenient, but at the same time, security problems are more serious, authentication must be more safely and effectively in order to ensure people’s property and personal safety. Because of the low cost of the technology required, the convenience of features extraction and the advantage of expansion, speaker recognition can be applied to a number of areas, such as computer remote login, security authentication, access control systems and other fields. In recent years, speaker recognition has become more and more popular. In numerous method of speaker recognition, this paper focuses on Mel-Frequency Cepstrum Coefficients (MFCC) and Gaussian Mixture Model-Universal Background Model (GMM-UBM).Those are the main research content of this paper:(1) We introduce the research purpose, background and significance of speaker recognition, analyze its development of the history and current situation both at home and abroad, and analyze the situation in used of practical application;(2) We explain the concept, classification and process of speaker recognition, analyze several method to extract speaker features and method of identification, and give the evaluation method of the system;(3) We introduce the pretreatment process before feature extraction, and analyze MFCC feature, dynamic information and short normalized extracting energy; and verify the effect of dynamic information and short normalized extracting energy;(4) We focus on the implementation of the GMM-UBM model, analyze the EM algorithm and so on, and compare various parameters of the model, finally obtain optimal results.