The Development Of Speaker Recognition System Based On Support Vector Machine

At present, Speaker Recognition has become a hotspot in authentication identification and artificial intelligence areas, has been widely used in our lives. SVM(Support Vector Machine) is a new and effective machine leaning method, which is advanced on the base of statistics theory. SVM can resolve a series Problems, such as non-linear, over-learn, high-dimension, local minimum value. SVM has strong generalization ability. This article is based on support vector machine theory and method and research on how to construct the speaker recognition system.In the process of building speech recognition system which is based on support vector machine, we studied some issues such as establishing the speaker speech database, speech signal pre-processing, speech endpoint detection, comparison of different parameters characteristic vector analysis, support vector machine parameter determination and system construction. The establishment of a standardized small speech database which is based on SQLite database, the organization and completion of the work of the speech samples collected is laid for the algorithm testing. Compared with the advantages and disadvantages of Mel Frequency Cepstrum Coefficient(MFCC) and Linear Prediction Cepstrum Coefficent (LPCC), MFCC parameters which reflects the static characteristics of speech signal has been used as characteristic parameters of speech signals. According to the basic speech signal processing method and the basic principles of speaker recognition speech recognition system was constructed after determining the parameters. In the process of establishing the speaker recognition system, we encountered some difficulties, such as the selection of feature vector frames, frame length and the selection of kernel functions and parameters that affect the recognition rate and time.Analysis experiment has been carried out. Through practical training and recognition speech recognition performance verification, optimization of parameters can be achieved 98% recognition rate.The system based on the computer hardware will can fully use the LabVIEW and MATLAB to construct speaker recognition system.