Research And Improvement Of Speaker Recognition Technology

Speaker recognition technology is a hot topic which has a wide range of applicationsin such areas such as security, public security and judicial, military, financial andinformation services.In the pure voice environment, the e?ect of recognition is good.Butin noisy environment, the recognition rate is relatively low.Considering this situation,the author embarked on a series of studies.The main activities include the following:1. Analyzed of the current development of speaker recognition,the technical dif-ficulties and the need to improve on. And researched the associated content of thespeech signal processing.2. Discussed two speaker recognition algorithms: Vector Quantization (VQ) andGaussian Mixture Model (GMM).Established a text-independent speaker recognitionsystem with the two methods, and simulated the two methods with MATLAB.3. In light of the insu?cient of feature extraction of speaker recognition, the authorresearched a method of transforming the characteristics after MFCC with principalcomponent analysis (PCA), Linear Discriminant Analysis based on the Fisher (FLDA)and PCA + FLDA. Also studied the improving speaker characteristics with wavelettransform method.The research shows that the methods can achieve good results.4. Speech enhancement techniques based on microphone array are analyzed.Delay-sum beamforming, Adaptive beamforming and Microphone array speech enhancementby signal subspace approach are analyzed in detail.And researched the method of speechenhancement with wavelet transform, and put wavelet transform methods after adap-tive beamforming. Finally, the various methods were simulated.5. Microphone array speech enhancement and feature extraction with PCA +FLDA method were used for speaker recognition in low SNR, which greatly improvedthe speaker recognition rate in noisy environment.