Kovtun V. Information technologies for increasing the dependability of the information systems for critical use with person authentication by voice

In the thesis for the first time: - the information technology of person-user authentication by voice in a speech signal is proposed in which, unlike existing, the speech signal presented by modulating and polyharmonic carrier fluctuations is parametrized in space of such characteristic parameters as average pitch frequency and amplitudes of carrier fluctuations in the system of quasi-deterministic and stochastic models of voice individuality; - A method for estimating the level of the signal-to-noise ratio in an empirical speech signal is proposed in which, unlike the existing ones, the last is considered as the sum of modulating and polyharmonic carrier fluctuations and white noise parameterized in the formalism of a refined quasi-deterministic voice individuality model; - A method for calculating the measure of the distance between the weighted characteristic parameters of the reference and empirical speech oscillations, parameterized, unlike the existing ones, in the formalism of the methodology of voice individuality models, where when estimating the confidence intervals for varying the values of the characteristic parameters of individuality of speech signals, the level of the signal-to-noise ratio is taken into account; - The information technology for optimization of the chosen attribute of dependability of a target information system for critical use is offered, for the description of which, unlike existing, the methodology of Markov, semi-Markov and managed semi-Markov models of confidentiality, integrity, availability, functional security, survivability, reliability are synthesized; - Methods of research of interdependence "level of confidentiality-to-integrity" and "level of confidentiality-to-readiness" are proposed in which, unlike existing ones, for competing attributes of dependability capacity presented in formalism of profile semi-Markov mathematical models, one-criterion optimization tasks are carried out are offered. Also improved: - The model of the process of person-user authentication by the machine of reference super- and i-vectors according to the parameterized representation of speech material with noise is processed; - Method of adaptation of an convolutional neural network for person-user authentication by speech material with noise.