A systematic literature review on Generative Adversarial Networks (GANs) based audio steganography

Steganography involves hiding secret information within digital media files, ensuring secure communication while maintaining its quality. Generative Adversarial Networks (GANs) have emerged as a transformative technology in this field. Existing reviews are mostly focused on either GAN or steganography. This article systematically reviews advancements in GAN-based steganography. 128 research papers from five leading scientific databases, including IEEE Xplore, Elsevier, Springer, ArXiv, and Wiley, were systematically analyzed in the first stage using a structured review process, focusing on GAN-based steganography methods to identify trends in GAN adoption in steganography work. The review covered publications from 2019 through April 2025. The corpus of 128 articles comprises peer-reviewed journal and conference papers and high-impact ArXiv preprints that demonstrated significant methodological innovations. Selected pre-2019 works are cited for contextual background only. Next, 28 articles were reviewed in detail on GAN-based audio steganography works only to identify methods, embedding domain, dataset, challenges addressed, and limitations of the works. Results show a significant increase in research adopting GANs. Standard GANs account for the largest share of implementations (55%), owing to their effectiveness in maintaining natural audio quality while embedding concealed data. Datasets like LibriSpeech, synthetic audio datasets, and Deep Sound Songs use the WAV format, which is the most popular audio format in studies. Although the use of traditional embedding methods is not mandatory, hybrid approaches commonly integrate Least Significant Bit techniques to achieve controlled and predictable embedding behavior. Developments like the inclusion of GANs in embedding schemes and cross-media steganography have enhanced robustness. However, issues such as restrictions on real-time performance persist. These findings suggest that future research should focus on computational speed optimization and the development of task-specific GAN architectures to make GAN-based audio steganography more practically viable.