Speech Recognition Access Control Applications

Speech recognition applications in access control is to make computers understand human voices and determine the speaker's identity. Speaker recognition is a biometric identification technology, it is a voice waveform according to the speaker will reflect the physiological and behavioral characteristics of voice parameters, it is automatic identification technology. Difference with the current speech recognition is to getting speaker information from the voice message, regardless of the meaning of words in the voice, which emphasizes the speaker's personality; but speech recognition aims to identify signal content from the voice , not considering who is talking, it emphasizes the similarities. Speaker recognition technology development due to the improve of information detection and processing, artificial intelligence, pattern recognition, machine learning techniques and theory development, which is a matter of physiology, phonetics, psychology, acoustics and other multi-disciplinary field of study.Speaker recognition technology is an important direction of development of computer technology.Speaker recognition has formed a complete theoretical system, Speaker recognition system in PC is obtain some success, although the speaker recognition research project has entered the commercialization stage, the basic theory quite well, a variety of products have emerged, but in many specific areas, because of its special nature of the industry and the environment, programmers often need to be developed specifically to meet the various industries actual needs.This paper describes an overview of speaker recognition technology and development at domestic and foreign, clarify the background and significance, analysis of the speaker recognition is facing difficulties. Second, according to the speaker recognition system basic structure describes the pre-processing, endpoint detection and feature extraction, including Linear Prediction Cepstrum and Mel Frequency Cepstral Coefficient(MFCC), and then introduced to the speaker template generation and template management and template matching and these related voice digital signal processing, pattern recognition and other basic principles, introduced on the speaker recognition system in the realization of several ways. Then, this thesis describes the use of Matlab program MFCC (Mel Frequency Cepstral Coefficient) and VQ (Vector Quantization) model's algorithm, and the Windows in use. Net, C # implementation of a simple text_independent distributed voiceprint attendance system, analysis of the vector quantization recognition rate and error reasons .Finally, use current popular programming technology and database technology to complete a speaker products.Because VQ model is only suitable for small-scale people of speaker recognition, in the case of identifying the growth of the recognition rate will be reduced, so I introduces another kind of speaker recognition model - GMM(Gaussian Mixture Model), This is a probability-based statistical model, even in the growing number of people , the recognition rate remains relatively stable. System improvements for future consideration, the paper also describes the HMM (Hidden Markov Model), which is also a probability-based statistical model, but this model is not used to identify who the speaker, it is to identify the content of speech . System improvements will get a great help by these models in future.Finally, a summary of the topics that the direction of improvement of speaker recognition. This article through to the actual speech recognition system of testing and research, for further development of practical speech recognition system work foundation and exploratory work.