An adaptive scheme for real-time audio authentication

Recently, there has been a significant increase in audio forgery, impersonation, and fabricated audio recordings due to widespread increases in software resources. Several techniques have thus been proposed to detect these risks, yet most of these techniques adopt processes that do not take into account the needs of real-time communication. Moreover, they rely on authentication codes being added to the transmitted signals, which leads to an increase in transmitted data size. In this paper, an automatic system for detecting fraud and impersonation is proposed that is suitable for both real-time communication and audio recordings, being characterised by simplicity and high processing speeds. The proposed scheme relies on the lifting wavelet transformation technique to implement human auditory signals across certain frequency bands to simulate the acoustic system, thus hiding an authentication code in the imperceptible bands. As a result, the original audio maintains its quality without an increase in the size of the transmitted signal. The input audio is divided into short periods (16 msec), with the authentication code adaptively extracted from an input frame and embedded in the next frame. The quality of the output signal thus maintains high quality, with at least 40 dB of SNR.