End‐to‐end audio steganalysis and payload classification using SSAM and statistical feature fusion

Summary The rise of social media has increased the risk of cyber‐attacks through covert communication channels. Steganography, originally used for secure military communications, is now exploited by hackers, threatening society. Therefore, effective detection of hidden information is vital for forensic investigators. This study proposes a novel end‐to‐end audio steganalysis framework consisting of hidden message detection, extraction, and identification of the extracted message phases. Initially, blind audio analysis, regardless of the steganography method used, utilizes a subtractive sample adjacency matrix and ant colony optimization to enhance classifier performance. The second phase involves a decoder that accurately recovers embedded payloads. Finally, the payload is classified as benign or malicious using a rule‐based heuristic and a random forest classifier. Experimental results indicate that the proposed system achieves an overall average detection accuracy of 97.15% across two benchmark datasets for stego audio and demonstrates strong performance across various embedding rates. The proposed framework provides a comprehensive solution for the audio steganalysis and has the potential to assist forensic investigators in detecting covert communication and strengthening cybercrime investigations.