Detecting double compressed AMR audio using deep learning
Da Luo, Rui Yang, Jiwu Huang · 2014
The Adaptive Multi-Rate (AMR) audio codec is a widely used audio data compression scheme optimized for speech and adopted by many devices. With the audio editing software, it is easy to perform tampering on digital speech recording, which makes the audio forensics become an important and urgent issue. Usually, the tampered AMR audio is double compressed AMR audio. In this paper, we proposed a method to detect the double compressed AMR audio. Such technique may be served as a tool for authenticating the originality of audio recordings and detecting the forgery positions. Our proposed method is based on deep learning algorithm and a majority voting strategy is designed for decision. The experimental results show that our method is effective to detect the double compressed AMR audio. Besides, the potential application of this technique is also discussed.