Double Order Hybrid Optimum Codebook For Voiceprint Recognition System

Voiceprint identification is one of most important biometric technology. It isidentified by characteristic parameters extracted from the speech signal, which based onspeaker’s pronunciation mechanism and pronunciation habits. It is widely used in manyfields, including biomedical, judicial expertise, network security, speech retrieval etc.Voiceprint identification is the focus of speech signal processing, which is graduallydeveloped from laboratory research into production research. In voiceprint recognition,the quality of codebook is very important, and the main research direction is to fastrealize the accurate codebook in the future. This paper research the preprocessing ofvoiceprint identification-codebook design, the performance of recognition algorithm etc,which is supported by Dazhou major scientific and technical project, and aimed topropose a novel and effective voiceprint recognition technology.In this paper, firstly, the basic principle of voiceprint identification has beenstudied in detail, and then, the method of analysis and processing of each part are putforward. These methods include analysis and processing of time domain and transformdomain of speech signal, principle and process of different characteristic parameters,and approaches of pattern match. Secondly, this paper studied the classic algorithm ofvector quantization, namely LBG algorithm, which is applied to voiceprintidentification. Vector quantization has some drawbacks, such as easy to fall into localoptimization, more sensitive to initial codebook, bad adaptive, so we introduce theglobal optimization algorithm-genetic algorithm (GA). The GA is applied into thepractical application of voiceprint recognition. After that， this paper proposed a novelvoiceprint recognition algorithm which is double order hybrid optimum codebookdesign algorithm by combining the two algorithm. Finally，the algorithm performanceof this paper is deeply analyzed under different parameters, including multiplecharacteristic parameters, the size of codebook, the length of speech, the value ofdistortion threshold, the dimension of characteristic parameters.The structure of this paper is as follows: chapter one introduce the back ground,significance and relevant research present situation, and relevant the current researchsituation; chapter two illustrates theoretical basis of voiceprint recognition system,including three part which contains pre-processing of speech signal, features extractionand pattern matching; double order hybrid optimum codebook design algorithm is given in chapter three, including relevant theoretical analysis bases on vector quantization andgenetic algorithm, the realization of the algorithm and experimental results; chapter fourdo research on different parameters on the system’s performance; chapter fivesummarizes the major contribution of the paper and the future work.