Research Of A Feature Extraction In Speaker Recognition

Speaker recognition technology is a comprehensive research project which uses psychology, physiology, digital signal processing, pattern recognition and artificial intelligence. It is the processing of automatically recognizing who is speaking by using speaker specific information included in speech signal. What kind of voice messages can effectively characterize a person and distinguish themselves from others. It is a very important part in the speaker recognition technology: Feature Extraction. At present, Mel cepstrum, LPC cepstrum coefficient, and kinds of coefficients that derived from spectrum are used in speaker recognition. But just only using one parameter can not achieve a good recognizing effect. Research has shown that more than one parameter combination can improve the recognition rate. Now, most combinations of parameters directly superimposed the other parameters behind a parameter. Although this method can improve the recognition rate, but increase a great deal of calculation. Based on the comprehensive studying of the linear prediction coefficient(LPC), the MFCC and their respective advantages, disadvantages, this paper gets a new feature extraction method. This article obtains a new feature factor which is called linear prediction Mel frequency cepstrum coefficient based on the linear prediction coefficient and Mel frequency cepstrum coefficient. New feature factor is not only reflecting the characteristics of a speaker, but also has a better robust. The new characteristic parameter is not the simple superposition of two parameters together, but integrate organic. Comparing to the characteristics of these simple combinations of parameters, Computation is much less. In order to verify the validity of the new characteristic parameters, this paper implements a simple speaker identification system using the Matlab7.0. We tested the recognition rate of LPMFCC, MFCC and the feature based on wavelet transform. The experiment suggests that the new feature factor is not only effective, but also achieves high recognition rate.