POTENTIAL INTELLIGIBILITY, A NEW INDICATOR OF SPEECH SECURITY

Research objective: to develop approaches for assessing speech intelligibility in confidential negotiation rooms, considering the intruder's capabilities in noise reduction and reconstruction of speech signals distorted by acoustic protection measures. Research methods: digital signal and image processing, digital time-frequency analysis, image-based speech analysissynthesis, comprehensive theoretical and applied analysis of experimental results on improving intelligibility of speech distorted by noise and interference. Results: limitations of existing methods for assessing the security of negotiation rooms against speech information leakage have been analyzed, which do not account for the capabilities of modern intelligibility restoration technologies. It is shown that the potential intelligibility index adequately reflects the real abilities of a skilled intruder to extract semantics from noisy speech messages and can serve as a new indicator of speech information security in negotiation rooms. Scientific novelty: new approaches to assessing the security of negotiation rooms are proposed, based on a comparative analysis of spectrogram images of the original and reconstructed signals. This aligns with the fact that the focus of research interest, both domestically and abroad, is actively shifting from classical intelligibility assessment metrics to models for its evaluation based on spectrogram images using deep learning methods and tools. Practical significance: the obtained results will allow for the expansion of capabilities of existing speech information protection systems in negotiation rooms and the development of more effective means based on the presented approaches.