An Approach to Assessing the Security of Speech Acoustic Information Using Neural Networks
Nikita A. Volkov, Andrey V. Ivanov · The Herald of the Siberian State University of Telecommunications and Information Science · 2024
The paper is devoted to the consideration of the methodology for assessing the security of speech acoustic information in the preparation of premises for private negotiations. Taking into account the disadvantages of existing approaches it is proposed to apply recognition methods based on convolutional neural networks. The paper proposes a block diagram of the stages for creating an intelligent system. The process of creating a training dataset in audio recording format with superimposed noises with different signal-to-noise ratios is described. The possibilities of the Adobe Audition audio editor and Python libraries for generating datasets are considered. It is proposed to classify spectrograms or mel-frequency cepstral coefficients of audio recordings using a neural network by the percentage of speech intelligibility in order to automate the process of assessing the security of speech acoustic information. To achieve the desired result, it is planned to train a neural network on various data, conduct a comparative analysis with the existing approach, evaluate the performance of the system and validate the results. The proposed approach and its practical application will significantly improve the quality and expand the conditions for the application of the security assessment of speech acoustic information.