Robust Anti-forensics on Audio Forensics System

Audio forensics systems are effective evidence to prove whether audio is real or not. Recently, there have been studies on anti-forensics technology that can deceive audio forensics systems. However, the anti-forensics technology loses its effectiveness after being compressed. In this paper, we first study the influence of known and unknown compression on audio anti-forensics technology, and enhance the robustness of audio anti-forensics technology. In this paper, the compression algorithm and compression approximation algorithm are added to the iterative process of generating anti-forensics adversarial examples to generate anti-forensics adversarial examples that can resist compression, and solve the problem that audio anti-forensics adversarial examples cannot reduce the accuracy of audio forensics system under compression. Three compression algorithms, AAC-HE, AAC-LC and MP3, as well as a pre-trained compression approximation algorithm, were added to the iterative process of generating audio anti-forensics adversarial examples to generate audio anti-forensics adversarial examples, which were uploaded and downloaded through the actual network platform (Himalaya) and input into the audio forensics system. Experimental results show that the proposed method is effective in resisting compression.