Research On Voicepirnt Recognition System Based On MFCC

With the development of information technology and network communication, the wholeof human society into the information age, we need to face a very important issue—information security. Identity authentication is an important part of information security,which affects all areas of our lives. Voiceprint is the inner and unique feature of a person,which likes fingerprint, face, DNA, by extracting the feature of the voiceprint for identityauthentication technology continue to develop, attracting many domestic and foreign scholars’research. Voiceprint recognition technology is the key to the speech signal front-endpreprocessing, feature extraction, modeling methods and anti-noise and other issues. Thispaper makes the corresponding research against the issues.Firstly, in order to reduce interference noise signal and improve the SNR of the speechsignal, this paper studies the speech signal preprocessing of the voiceprint recognition,analysis the spectral subtraction speech enhancement algorithm, and the spectral subtraction isimproved, the focus is proposed based on recursive least squares (RLS) adaptive noisecancellation method, simulation results show that, the effect of RLS adaptive filtering speechenhancement is better.Then, this paper detailed studies the speech signal feature extraction of the voiceprintrecognition, describes the extraction methods of feature parameters Formant, LinearPrediction Cepstral Coefficients (LPCC), Mel Frequency Cepstral Coefficients (MFCC),focusing on analysis introduced the Gammatone Frequency Cepstral Coefficients(GFCC)based on the human ear cochlea model, proposed the extraction algorithm of featureparameter GFCC, realized the extraction of formant, LPCC, MFCC, GFCC by simulation.Secondly, in order to improve the robustness and the recognition rate of voiceprintrecognition system, this paper studies the Gaussian mixture model (GMM) parameterestimation expectation maximization (EM) algorithm and the initialization method, elaboratedGMM training and recognition methods. Further in-depth study the experiment effects of thevoiceprint recognition system based on GMM under the feature parameters in differentdimensions, mixed numbers in different models and different SNR environment by thesimulation experiments, comparing the performance of different dimensions featureparameters LPCC, MFCC, GFCC, the performance of different GMM mixed numbers. Inaddition, under different SNR, a voiceprint recognition algorithm based on GFCC and RLSadaptive filtering speech enhancement algorithm is presented. Experimental results show that,the GFCC has anti-noise capability, can improve the intelligibility and clarity of the speechsignal, better than recognition performance and robustness of MFCC; and the voiceprintrecognition system which combined GFCC with RLS adaptive filtering, the recognition ratecan reach over90%, the recognition performance is superior to the combination of improvedspectrum subtraction and MFCC recognition algorithm.Finally, this paper analyzes the results of the study and makes a detailed description ofinadequacies, then proposes the future research directions.