Skip to content
Conference Open access

Multi-Class Classification of Different Types of Hidden Data in Audio Steganography Using CNN-Based Models

Sep 2026 · 2026 11th International Conference on Computer Science and Engineering (UBMK) · pp. 823-828 · 0 citations · 17 references

Abstract

In today’s digital era, data privacy and cybersecurity have become among the most critical components of digital communication. Steganography provides a secure means of information exchange by embedding secret data into digital media in a manner that is imperceptible to human perception. In this study, the limitations of the Least Significant Bit approach, one of the most widely used data hiding methods in audio steganography, are examined in terms of detectability and data representation. To address these limitations, an adaptive dataset generation framework was developed that supports different types of hidden data and dynamically adjusts the embedding process according to signal quality. In the proposed approach, speech signals in Waveform Audio File format were used as cover media, while image-, text-, and audio-based payloads were embedded using the Least Significant Bit method. Unlike conventional implementations, the embedding intensity was gradually increased to preserve perceptual quality while minimizing statistical distortions. Using the proposed Stego4Audio dataset, the performance of various Convolutional Neural Network-based architectures was evaluated on steganalysis tasks involving four distinct classes: Cover, ImageStego, TextStego, and AudioStego. Experimental results showed that DenseNet achieved the best performance, obtaining an F1-score of 98% in the binary classification task and 94,76% in the four-class classification task. The findings demonstrate that the proposed approach is effective not only for stego detection but also for identifying the type of embedded content.

Read PDF

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.