CEPSTRAL PITCH DETERMINATION FOR VOICE RECOGNITION

摘要

One of the challenging tasks in an Automated Security System is to identify a human, uniquely. Recent years have seen an explosion in research efforts to put biometric traits for practical implementation of security system. Each human have his own characteristic features, which forms the base of the identification process. Voice is one such versatile biometric feature. The access of Automatic Teller Machine (ATM) using voice biometrics is proposed in this paper. "Pitch" effectively characterizes the biometric features of human voice. A powerful speech signal-processing tool called the "Cepstrum" analysis derived from "Homomorphic" process, due to its 'deconvolution effects' is used as a "pitch-detector". Cepstrum is defined as the power spectrum of the logarithm power spectrum. The computational efficiency of the FFT is being exploited for the calculating the power spectrum twice. "Vector Quantization" is used as a pattern recognition system. From the quantized values, the Error Distances are calculated which are then compared with the threshold distance values to determine whether the identity claim should be accepted or rejected for accessing the ATM. The process is simulated using a simple MATLAB program and the simulation results are plotted.