Title: Towards breaking DNN-based audio steganalysis with GAN

Authors: Jie Wang; Rangding Wang; Li Dong; Diqun Yan; Xueyuan Zhang; Yuzhen Lin

Addresses: College of Information Science and Engineering, Ningbo University, Ningbo, 315211, China ' College of Information Science and Engineering, Ningbo University, Ningbo, 315211, China ' College of Information Science and Engineering, Ningbo University, Ningbo, 315211, China ' College of Information Science and Engineering, Ningbo University, Ningbo, 315211, China ' College of Information Science and Engineering, Ningbo University, Ningbo, 315211, China ' College of Information Science and Engineering, Ningbo University, Ningbo, 315211, China

Abstract: Recently, deep neural network (DNN) has significantly boosted the performance of audio steganalysis. Accordingly, most of the traditional steganography cannot resist such DNN-based steganalysis. In this work, we attempt to break a given DNN-based audio steganalysis method with a prespecified steganography. To achieve this goal, we propose to employ a generative adversarial network (GAN) to generate an enhanced cover audio firstly, which can be regarded as more suitable for steganography. Then, the secret message is embedded into the enhanced cover audio with the traditional steganography, instead of into the original cover audio. The experimental results demonstrate that our generated enhanced cover audio could effectively aid traditional steganography to break the advanced DNN-based audio steganalysis.

Keywords: steganography; GAN; generative adversarial network; adversarial examples; deep learning; steganalysis; deep neural network; audio signal; cover enhancement.

DOI: 10.1504/IJAACS.2021.10030239

International Journal of Autonomous and Adaptive Communications Systems, 2021 Vol.14 No.4, pp.371 - 383

Received: 27 Nov 2019
Accepted: 23 Apr 2020

Published online: 23 Nov 2021 *

Full-text access for editors Access for subscribers Purchase this article Comment on this article