Research & Design Of Identity Recognition System Based On Voice

The identity recognition based on voice is the processing of automatically recognition whether the speaker in the speakers whose voice records has been gathered, then determine who the speaker is, by analyzing the speaker's pronunciation signals and picking up the speaker's characteristic. With the development of computer technology and information society, the speaker recognition technology receives more and more attention, and it has good application prospects in many fields. By analyzing the general principles and system structure of speaker recognition and considering subsistent technology of speaker recognition, Linear prediction cepstrum coefficient and Mel cepstrum coefficient are adopted as characteristic parameters, the vector quantization is used as speaker recognition method to set up speaker recognition system. Currently in the field of speaker recognition, there are two important questions need to solve for the enhancement of recognition rate. One is how to select more effective and more reliable speaker's characteristic, the other is how to select the best recognition model and pattern analysis methods. This article has made the discussion for this, and made following aspects research and improvement.(1) In this paper, I analyze the common principle of the voice signal, and focus on the research about the three sub-model of digital model of voice signal: inspired model, sound gate model, and radiant model.(2) Based on the principle of feature parameter, I study the parameters representing the voice signal of the speaker. Two kinds of representative features, LPCC and MFCC, are analyzed and obtained. The performances of LPCC and MFCC are compared respectively on computer platform. Experiments indicate that the systems can gain better performances adopting MFCC than adopting LPCC on computer platform, and the higher probability of recognition is obtained with more training time.(3) I explain the principle of some primary modeling methods about speaker recognition, and VQ is emphasized, also illuminate the best algorithm to build a speaker-unspecific VQ codebook—LBG algorithm. I design and realize the system of speaker recognition with VQ model on MATLAB platform and give the detail test results and analysis. Finally, aiming at the deficiency of VQ, We put forward the improved method for clustering. We also introduce and realize a kernel method for pattern analysis named novel detection.The noise robustness is one of the key problems for the practicability of speaker recognition system. How to keep performance of the systems not dropping in a situation that the noise environment changes, further research and practice is necessary. With the development of relevant subjects, some more practical and more high-powered speaker recognition system will appear and apply to people's actual life extensively.