ASSESSMENT OF THE SECURITY LEVEL OF HEAVILY NOISE-DISTORTED SPEECH INFORMATION BASED ON THE RESIDUAL INTELLIGIBILITY CRITERION

Abstract

This paper substantiates the scientific foundations for assessing the security level of heavily noise-distorted speech information in technical channels of information leakage using the residual intelligibility criterion. The limitations of traditional approaches to speech-security assessment are analyzed, including SNR, SPC, AI, and SII metrics, which do not account for nonlinear distortions, the frequency-selective properties of the environment, or acoustic reverberation effects. An integral metric, RII, is proposed to characterize an adversary’s ability to reconstruct the informative content of a signal after interference exposure. A generalized speech-processing model is developed, combining wavelet filtering, segmentation, triphone-structure analysis, and neural residual-intelligibility estimation. A classification of major noise types is presented, including white, reverberant, combinational, and speech-like noise, along with approaches to their mathematical generation. It is demonstrated that incorporating the time–frequency characteristics of the environment and the speaker’s individual features increases the accuracy of modeling hazardous scenarios of speech-information interception. The proposed approach enables objective ranking of leakage channels and supports rational decision-making regarding active and passive acoustic protection. The methodology can be applied to the design of comprehensive technical information-protection systems, threat expert analysis, and evaluation of the effectiveness of speech-suppression measures under complex acoustic conditions. Special attention is given to modeling the acoustic interaction between the source, enclosing structures, and technical reconnaissance devices, enabling quantitative assessment of the probability of partial or complete reconstruction of message content. The developed methodology accounts for phase distortions, structural spectral maxima, adaptive-filtering characteristics, and conditions of tonal masking. From the standpoint of reducing residual speech informativeness, a comparative analysis procedure is proposed for evaluating the effectiveness of different types of interference.