Research And Application Of Voiceprint Recognition In Access Control System

Voiceprint is an important biological feature.Voiceprint recognition technology is also called the speaker recognition.Access control system based on voiceprint recognition is safe,reliable and easy to use.It is a hot research topic at present.This thesis takes voiceprint recognition in the access control system as the research problem.On the basis of explaining the basic structure and work flow of access control system,design and implementation of the voiceprint recognition module for the access control system is given in detail.Voiceprint recognition in access control system can be divided into two parts:voiceprint registration stage and voiceprint recognition stage.Voiceprint registration mainly completes data acquisition,preprocessing and model building for speakers.In the voiceprint recognition stage,the speaker is detected according to the speaker voice pattern obtained by the training.The core of the algorithm is the establishment of speaker recognition model in voiceprint registration.In order to effectively obtain features from speaker audio data,In this thesis,speech signal is preprocessed by pre-emphasis,frame division and window addition,converting speech signals from time domain to frequency domain by Fourier transform.Considering that the MFCC does not adequately reflect the problem of voice dynamics characteristics,apply multi-window spectrum estimation and first-order second-order difference to feature extraction process,and a modified MFCC spectrogram is given.On this basis,the thesis uses the acquired spectrogram as the data source,Convolutional Neural Network model for establishing voiceprint recognition.The validity of the model is verified on TIMIT and self-built data sets.The experimental results show that the voiceprint recognition accuracy will be greatly reduced in noisy environments.To solve this problem,the voiceprint recognition with noise is divided into three processes in this thesis: noise type recognition,noise removal,and voiceprint recognition.The recognition of noise type is realized by using BP neural network model respectively.LSTM model is used to denoiseing.Voiceprint recognition is accomplished by convolution neural network.Finally,the effectiveness of the proposed scheme is verified by experiments.