About the possibility of using speech recognition technologies in problems of assessing the protection of acoustic information from the leakage throught technical channels

Abstract

This article discusses the possibility of improving the existing instrumental calculation method for assessing the security of meeting rooms by moving from the Pokrovsky formant method to using speech recognition algorithms based on deep neural networks. In our case, it is proposed to use recurrent neural networks that have shown themselves best in terms of information processing with low SNR ratios. Also, using the long-term short-term memory of a recurrent neural network, it is proposed to improve the test signal. In addition, it substantiates the need to take into account the individual characteristics of the room and the possibility of using interference, such as a “speech choir”.

Development of a secure video chat based on the WebRTC standard for video conferencing

“…WebRTC uses two audio codecs, G.711 and Opus, as well as VP8 and H.264 video codecs. In most cases, developers prefer to use Opus audio codec and VP8 video codec due to their characteristics [10][11][12][13][14][15].…”