Research On Voiceprint Recognition Technology And Application Based On MT MFCC And Improved V Neural Network

With the promotion of mobile Internet of Things and artificial intelligence in recent years,the idea of empowering all walks of life under the premise of ensuring technical standardization has become more urgent.Among them,voiceprint recognition is a technology that automatically finds a match and recognizes the speaker's identity from the speaker's real-time voice or recording.Voiceprint(speaker)recognition technology continues to develop and has become an economical and reliable method of identity recognition and verification.In recent years,the introduction of artificial neural networks with the characteristics of autonomous learning ability,nonlinear fitting ability,and associative storage function has provided new ideas and methods for voiceprint(speaker)recognition,a complex nonlinear process.In this thesis,based on the Back Propagation(BP)neural network,further study on voiceprint(speaker)recognition technology and application.First of all,this thesis gives an overview of the principles and methods of voiceprint recognition technology.Looking at the current status of the technology in various fields and the functional requirements of laboratory projects,in order to solve the problems of low recognition accuracy and slow speed in noisy environments,In-depth research on the two key technologies of feature parameter extraction and improvement of modeling methods,based on the acquisition of Mel-Frequency Ceptral Coefficients(MFCC),an improved multi-window spectrum MFCC feature parameter(Multitaper MFCC,MT MFCC)extraction method is proposed.Secondly,through preprocessing such as voice data collection,pre-emphasis,framed windowing,endpoint detection,and signal enhancement,this article eliminates the influence of human factors and equipment reasons on the quality of voice signals.The MT MFCC feature parameter extraction method is used to emphasize the low-frequency information of the voice to highlight the voiceprint recognition features and reduce the parameter fluctuations,which effectively solves the problem of large variance when the traditional feature parameter extraction method is used in the voiceprint recognition process.Finally,complete the overall scheme design,select the parameters to measure the performance indicators,construct the BP neural network voiceprint recognition model,propose genetic algorithm(Genetic Algorithms,GA)optimization,and increase the momentum factor to the traditional BP neural network algorithm and structure.Improve;the improved model is analyzed under different network structures and data distributions,and it is concluded that the setting of different parameters has a great influence on the recognition results.It is necessary to select appropriate numerical results through multiple experiments;the voiceprint recognition technology is applied to the voice recognition system of the speech robot "Yujiao No.3" platform,and the effect test of the voiceprint recognition model before and after the improvement under different noises is conducted to recognize different speakers.