Transformer for authenticating the source microphone in digital audio forensics

Abstract

In recent years, digital audio has played a significant role in providing information used as evidence in criminal investigations. The authenticity of an audio recording must be proven for it to be admitted into evidence in a legal proceeding. In this paper, we propose a robust system based on transformer deep learning architecture for identifying the source microphone with high accuracy, which is applicable in digital audio forensics. To evaluate the proposed model, we use the Audio Forensic Dataset for Digital Multimedia Forensics (AF-DB) and the King Saud University speech database (KSU-DB) datasets. Experiments show that the proposed model achieves state-of-the-art classification accuracy on the two databases. The model achieves an overall accuracy of 94.18% and 84.19% for Inter-model and Intra-model microphone classification, respectively, for the AF-DB and 99.38% for the KSU-DB speech dataset.