Research And Development Of Speaker Recognition System

As an important part of speech signal processing technology, speaker recognition has a very bright future for both research and applications. At present, there have several recognition methods applied in speaker recognition but recognition rate, system stability and noise robustness is not better. This paper presents the new method that the support vector machine technology has applied in speaker recognition. The method has obtain the higher recognition rate and the better noise robustness compared with traditional methods. Support vector machine technology is one of statistics learning theories,combining the equalization point of experience risk and expectation risk. It is a kind of newer machine learning method to give consideration to the learning ability and extending ability. It has a very great advantage while dealing the samples with nonlinear and multidimensional problems on the basis of the mode categorized method of the structure risk with minimizing principle. So there are good results on speaker recognition based on speech signal samples. In accordance with the actual project, this paper fully considers the comprehensive factors of speaker recognition system on the basis of explaining speaker recognition's principle, such as recognition rate, recognition speed, system stability etc. LPCC is used as the feature parameter. LPCC can reflect the esthesia feature of human in pronunciation and has offered a group of succinct speech signal model parameters which signify the range of frequency spectrum of the speech signal more accurately and the operation amount is not comparatively big. Combined vector quantization technology, the speech searched system has been developed. Vector quantization technology differents from traditional mark quantization methods, basic thought to unite several scalar data into a vector and carry on space divided to vector space, which offered quantization wholly, thus data are compressed in high quality. Because of the rigor of theory, simpleness, convenience and better design effect in LBG algorithm, LBG algorithm is adopted in clustering feature codebooks. It has guaranteed the quality of codebooks and systematic quality is got the assurance too. Against some practical problems of lower recognition rate and system instability of speaker recognition system under the noise environment, this paper analyses the recognition capability for each heft of feature parameters with the distinguishable functions through the experiment, and also analyses the speaker recognition under the different noise environment through the experiments. Those offer the strong data and positive suggestions for the studies of speaker recognition.