Contributed to a team research project simulating 16 steganography attacks across 8 devices and 4 distinct attacker profiles (“Alices”), each tracked under a unique case index (e.g. A1.1). Stego-audio samples were generated and analyzed using FFMPEG, SOX, PyTai, Sherlock, and an MP3 file structure analyzer, with results consolidated into a shared Excel sheet to surface patterns across samples.
My contribution was the automation layer: I developed Python scripts to automatically extract and structure data from the forensic tools used in the analysis, in particular Sherlock and the MP3 file structure analyzer, streamlining what would otherwise have been manual result collection across dozens of samples. I applied spectrogram and metadata analysis to detect embedding artifacts such as high-frequency distortions, and helped evaluate results against three forensic criteria: detectability of the attack (EDE), attribution to number of attackers (EAT), and traceability to a specific attacker (EDI). I also tested how well hidden payloads survived real-world transformations like recompression and format conversion.
The project ultimately identified consistent forensic signatures in manipulated audio, demonstrating both the viability of audio as a covert communication channel and the practical limits of its concealment under structured forensic analysis.