Research On The Algorithm Of Voiceprint Recognition Under Noise Background

With the continuous development of voiceprint recognition technology,voiceprint recognition has emerged a new situation of rapid development in recent years.With its convenience,stability and low-cost characteristics,voiceprint recognition has been widely used in military safety,medical health,judicial identification and other fields,and gradually become one of the mainstream biometric technology.Although in the noiseless environment,voiceprint recognition has achieved very good results,but in the practical application process,because of the influence of various environmental noise,voiceprint recognition often fails to achieve the ideal effect.Therefore,it is of great significance to study the technology of voiceprint recognition in the background of noise.Through the analysis and research of the whole voiceprint recognition system,the work and innovation of this thesis mainly include the following aspects:The feature parameter extraction algorithm of voiceprint is studied.By analyzing the GFCC feature extraction algorithm based on gamatone filter,this thesis improves and optimizes the feature extraction process of GFCC by using logarithm compression.In this thesis,an improved adaptive compression acgfcc feature extraction algorithm is proposed based on the GFCC feature extraction algorithm combined with the nonlinear compression characteristics of power function.The experiment and simulation are carried out under the framework of Gaussian mixture model_the general background model.The experiment shows that the average recognition rate of acgfcc features is 80% in 5dB noise environment,95% in 10 dB noise environment and 97% in noiseless environment.It improves the recognition rate of the system and has good anti noise performance.The anti noise performance of voiceprint recognition model is studied.Aiming at the problem that voiceprint recognition can't match well in training environment and practical application environment because of noise,this thesis optimized the parameters of voiceprint model,and proposed an algorithm of optimizing the parameters of voiceprint model based on spectral subtraction.The optimized voice model can be obtained by the test voice model subtracting the noise model in the logarithmic spectral domain.This method is simulated and verified in the frame of the Gauss mixture model_general background model.The experiment shows that the optimized voice model can improve the accuracy of the system by 1% to 2% under the background of noise,and to a certain extent,it can solve the mismatch between the training environment and the actual application environment with the background noise.