Analysis of methods for detecting artificially synthesized content

The paper presents an analysis of existing methods for detecting artificially synthesized content and proposes a proprietary architecture for DeepFake’s hybrid detection system based on searching original content. The study tests and compares the effectiveness of detection methods in two different cases. In the first case, records for training and testing samples are used from the same dataset; in the second case, testing is performed using a black-box method using records from different datasets. As a result, it is concluded that there are shortcomings in the existing methods and a hybrid DeepFake detection system architecture is proposed.