Audio Postprocessing Detection Based on Amplitude Cooccurrence Vector Feature

Abstract

Authentication of audio signals is an important problem in multimedia forensics. Tampering is typically followed by postprocessing that aims to remove the traces of the forgery. The variety of possible postprocessing operations makes tampering detection even more challenging. In this letter, we propose the amplitude cooccurrence vector features, which exploit cooccurrence patterns in audio signals. Experimental results show that our proposed features are able to distinguish between the original audio and the postprocessed audio with an average accuracy of above 95%. Furthermore, it can also effectively discriminate different kinds of postprocessing operations.