A group of staff and students from the Department of Cryptology and Cybersecurity at the National Research Nuclear University MEPhI, led by Candidate of Technical Sciences, Associate Professor Dmitry Efanov, has developed an intelligent system capable of evaluating a recording of a human voice to see if it is a voice deepfake. The project became the winner at the Moscow Mayor's Competition among innovative projects for the best entrepreneurs, scientists and inventors of the capital "Innovator of Moscow". The developed product was named "Barn Owl" - in honor of the nocturnal bird, characterized by particularly acute hearing.
According to Dmitry Efanov, the system is based on a neural network trained to distinguish deepfakes. More than 200 thousand recordings of human voices were used as data for training the neural network, including both authentic recordings and deepfakes generated by various algorithms. Deepfakes are detected by analyzing the so–called cepstral coefficients, which is what acoustics experts call some of the most important mathematical characteristics of audio signals, usually calculated for short fragments of audio recordings lasting from 20 to 40 microseconds.
According to the scientist, work on this system began in 2022, and then it was a purely academic task, since voice forgery at that time required a lot of resources and was not widespread. At the first stage, the project was supported by the Foundation for Assistance to the Development of Small Enterprises in the Scientific and Technical Field (the Foundation for the Promotion of Innovation).
However, by 2024, voice forgery technologies have become much more accessible and widespread – in particular, they have become widely used by fraudsters. Thus, the initially scientific development has prospects for applied use. Dmitry Efanov's group participated in the Technological Accelerator of MEPhI and Rosatom State Corporation – Barn Owl became one of the 4 finalists of the accelerator (out of the initial 60 applications).
Dmitry Efanov emphasizes that today the task of the Barn Owl developers is to create a cloud service that will allow you to easily and quickly connect the deepfake recognition function to any public or commercial communication system that needs it, for example, to a call center or messenger. Such well-known Russian companies as TBank and Informzashita have already confirmed their interest in using the service. In the future, it is also necessary to create a research laboratory in which the neural network will be improved on a regular basis, as new algorithms for synthesizing deepfakes are constantly emerging.
Scientists hope that the project will receive support from the relevant structures of the Moscow Government.
