Speaker Recognition System Based On Support Vector Machine

With the social progress and the rapid development of information processing, speaker recognition technology as a biometric technology has been widely used in our lives. This paper aims to design speaker identification system based on support vector machine .It is mainly made use of many unique advantages of SVM to settle the pattern recognition problems of limited samples, non-linear and high-dimensional problems in the performance .Researched SVM-based speaker recognition system, we can see from the principle of speaker recognition, speaker recognition system includes a speech signal acquisition, preprocessing, feature extraction, speaker model training and recognition part. The mostly part is the extraction of speech signal feature parameters and model selection and optimization algorithm, these factors will affect recognition rate of the system. As the speech signal is very complex, the characteristics of the selected parameters should be as complete as possible to accurately express the information of the voice signals. Based on existing advantages and disadvantages of the feature selection method, the selected feature method is MFCC of the simulation human auditory characteristics method. MFCC and the first-order differential parameters which reflect the static features and dynamic features of speech signals are assembled as the characteristic parameters of speech signals. In the pattern recognition algorithms, we use SVM model, there will improve recognition rate through the selected SVM's kernel function and optimized its parameters. As a result, the current optimized parameters can achieve recognition rate over 99%, false alarm rate about 2.4% and the recognition time less than 0.1s.The system based on the computer hardware will can fully use the advantages of LabVIEW of the graphical programming and flexible and MATLAB of the powerful data processing to build an interface of the SVM-based speaker recognition system for user identification. The system is easy to operate, to use in users, to achieve the acquisition of the identification and processing of the real-time signal.