Hiding images within audio using deep generative model

Subhajit Paul, Deepak Mishra

Abstract: Image steganography is a procedure of hiding any messages within an image. In this paper, our major goal is to conceal images within audio, and we converted this audio steganography problem to image steganography by utilizing the mel-spectrogram of the audio files as the cover medium. Previously, this audio steganography problem was implemented using statistical methods like least significant bit (LSB) encoding. Here we explore the use of deep neural networks (DNNs), and we propose a new technique to hide images within the audio using deep generative models which allow us to optimize the perceptual quality of the reconstructed audio and image by our model. We showed that our model efficiently hides images within audio and evades detection by steganalysis tools, is robust to different color spectrum images, and can hide multiple image data in single audio.