Research On Speaker Recognition In High Noise Circumstance

Voice print recognition, which is also called speaker recognition, is a technology that attempts to recognize a speaker through measurements of the specifically individual characteristics arising in speaker's voice. And it is different from voice recognition. Voice print recognition uses the speaker's information to identify the voice signal without considering the meaning of words in the voice and it stresses the speaker's personality. But voice recognition's aim is to identify the language in the speech signal and it does not consider the words. Voice recognition emphasizes the commonness.Recent years, in the biological identification technologies area, speaker recognition technology with unique advantages such as convenience, economy and accuracy is paid much attention to. Speaker recognition has a wide range of applications which include banking or credit card transactions by telephone, information and reservation services, access control in high security areas and forensic investigations. Though speaker recognition systems perform well when clean speech is used for training and testing, the performance degrades rapidly when speech used in real-world conditions.This article mainly studies the problems of speaker recognition system in high noise circumstance, including pretreatment, feature extraction and back-end processing and so on. Firstly, this paper uses the voice detection based on DCT and spectral entropy and speech enhancement based on spectral subtraction for the pretreatment of speech. This algorithm calculates rapidly and availably. It has good anti-noise ability. Secondly, this paper adopts feature extraction based on fast lifting wavelet transform and MFCC. The new feature parameters applying to the system has high recognition rate in low SNR. Thirdly, the part of back-end processing researches on speaker identification based on GMM and SVM.