Investigation of Informativeness and Stability of Mel-Frequency Cepstral Coefficients Estimates Based on Voice Signal Phase Data of Authentication System User
Mykola Pastushenko, Yana Krasnozheniuk, Maksym Zaika · 2020
The problems of increasing the reliability of using various resources, access to which is carried out by means of infocommunication networks, are considered. It is known that the first barrier in ensuring high reliability of access is a high-quality user authentication system. Currently preference is given to access systems based on biometric characteristics of a user. Initially, the priority was given to static biometric characteristics of a user (face image, finger papillary picture and iris). These biometric features did not meet the expectations of developers and users due to the simplicity of their forgery. At present, the main attention of developers is focused on implementation of dynamic (behavioral) biometric features of users and, first of all, voice authentication systems. Voice authentication systems have a number of significant advantages: simplicity, compactness, low cost, and a number of others. It is also important that the user's passphrase can be quickly changed and extended in the process of voice authentication. The quality indicators of voice authentication systems, like all biometric access systems, do not meet the increasing requirements. In the process of voice authentication, the amplitude-frequency spectrum of the recording materials is analyzed. The main efforts of researchers are focused on using estimates of the pitch frequency and formant frequencies associated with it, cepstral coefficients, mel-frequency cepstral coefficients, and linear prediction coefficients as a user template. Some attention is paid to decision-making procedures based on Gaussian Mixture Model, Support Vector Machine, and Hidden Markov Models or artificial neural networks. In the presented work, it is proposed to supplement the analysis of the amplitude-frequency spectrum with studies of phase data, which is currently receiving less focus in the process of voice authentication. The results of research on estimates of pitch frequency and mel-frequency cepstral coefficients based on the amplitude and phase information of the voice signal are presented. The purpose of this work is to analyze the informativeness of phase data of a voice signal, as well as to study the stability of estimates of the user's template and, first of all, mel-frequency cepstral coefficients calculated from the phase data. The studies performed have shown high informativeness and stability of the investigated estimates, which emphasizes the importance of the phase information of the voice signal for improving quality characteristics of voice authentication systems.