Implementation of secure speaker verification at web login page using Mel Frequency Cepstral Coefficient-Gaussian Mixture Model (MFCC-GMM)

The need of security for web page was increased as the development of online activity especially trading or banking. Speaker recognition can be used to secure the web page which need high security level. In this research, the speaker recognition system at web page was successfully built for login authentication security. For enrollment and verification need, speech signal from clients was recorded in 35 seconds for enrollment and 10 seconds for verification then transferred to server by network. Then this signal will be processed with Sampling, Frame Blocking, Windowing Hamming and Discrete Fourier Transform. The signal in frequency domain will be filtered by Nonlinear Power Spectral Subtraction to reduce background noise. For identification, the system extracts the feature of Mel Frequency Cepstral Coefficient (MFCC), and to build the model of these features uses Gaussian Mixture Model (GMM). To improve the security level, the system uses Secure Socket Layer (SSL) with 1024 bits RSA encryption. From this research, we have succeeded in optimizing the signal quality up to 5 dB SNR, the mean error recognition level of FAR is about 23.3% and FRR 27.5 and the maximum accuracy of recognition system is around 88% when the quality of speech signal is clean. The computation time for enrollment is about 552573,5 millisecond and for verification is about 129062,6 millisecond.