Research Of Voiceprint Recognition System Based On Dynamic And Optimal Codebook Designing

With the development and revolution of modern information technology andnetwork technology, information security has become one of the hottest researchareas in the world, and the speaker identity is one of the important parts.Additionally, with the advantage of small cost, non-contact and high reliability,voiceprint identification has gradually become the focus of biometric research.Currently, few published reports involve the research of voiceprint under differentchanging environments. In addition, voiceprint identification based on differentclasses of objects has not been reported publicly. Based on the correspondingresearch, this thesis expended the research within objects, codebook designing andrecognition algorithm of voiceprint recognition, and proposed the new algorithm ofmulti-channel parallel voiceprint recognition designing, dynamic class voiceprintrecognition and the implementation of system based on DSP. The results ofvoiceprint recognition are satisfying.The main work of this thesis is as follows:Firstly, the basic principles of vector quantization, genetic algorithm, BP neuralnetwork and Gaussian mixture models voiceprint identification algorithm have beenstudied in detail, and then, the multi-channel parallel voiceprint identificationalgorithm was designed to achieve optimal solutions. Meanwhile, the performance ofthe system was deeply analyzed based on different signal noise ratio, speechdurations, types of characteristic parameters and dimensions of characteristicparameters.Secondly, the concept of class identification and the dynamic optimalcodebook designing algorithm were proposed. The recognition of different types ofobjects could be achieved with the optimal class codebook.Thirdly, the design of the software and hardware system based on DSP wasproposed which is aimed to realize voiceprint recognition system with the embeddedDSP platform. This work could provide ideas to realize the hardware and softwaredesign of voiceprint recognition system.The achievements, including the proposing of multi-channel parallel voiceprintidentification algorithm, the designing of dynamic optimal codebook and therealization of DSP-based hardware and software of optimal voiceprint recognition system, provide theoretical basis and methods for the further study of the principle ofvoiceprint recognition and how to improve the performance of the system, and givesome reference ideas to promote the research and application of voiceprintidentification especially the class-oriented voiceprint identification.