A GAN-Based Framework for Text Hiding in Audio Files Using Spectral Transform Compression

Abstract

A Generative Adversarial Network (GAN)-based framework is a well-known method for concealing textual information within digital audio signals via spectral-domain embedding. Conventional audio steganography techniques, such as Least Significant Bit (LSB) substitution, are low in robustness to signal processing operations, including compression, filtering, and noise addition. To address these limitations, in this study, we embed text into the spectrogram representation of audio signals by using adversarial learning. Based on this model, it adaptively embeds the message in perceptually insignificant frequency regions, while the discriminator enforces statistical similarity between original and stego audio. This demonstrates high imperceptibility, robustness, and payload capacity compared to traditional methods.