Text-independent Speaker Recognition Technology

Speaker recognition technology has become a huge attraction and are both very difficult to research hot spots in recent years. Speaker recognition is issued from the speaker's voice signal in the automatic extraction of speaker information, and speaker identify areas of research. It is different from the speech recognition, which purpose is not to identify the content of speech,but to determine the identity of the speaker,or that is system judge to the speaker who is of the research.This paper is Text-independent Speaker Recognition Technology. The main research work is:Speech Database has been established in the voice samples for pre-processing work,including Pre-emphasis, Sub-frame, plus windows, endpoint detection. Focus on Dual-threshold method of endpoint detection. Reducing the amount of dataand the noise interference after pre-processing work. Follow-up work for the realization and analysis to lay a good foundation.Reserched traditional feature extraction(LPC,LPCC,MFCC) and improved feature extraction(LPMCC,3Q+l). Through experiments confirmed its feasibility,and then calculated separability-measure, after a preliminary verification of the algorithm to improve significantly superior to the traditional LPC, LPCC and MFCC.Application of HMM,VQ,ANN,SVM identify methods to identify the characteristic parameters.Constructed speaker recognition system.Experimental study on the adoption of further confirmed that the characteristic parameters of the improved recognition rate in the system is obviously superior to the traditional feature parameters.And comparatived analysis of four types of identification methods,Which ANN and SVM is better than HMM and VQ in identifying the effect.