The A-Files

Reproducible evaluation of audio steganography and watermarking.

The A-Files combines embedding methods, controlled channel distortions and objective audio measures in one experiment engine. Evaluate payload recovery, perceptual distortion, empirical capacity, robustness and statistical detectability on the same recordings and conditions.

Components	Documentation
30 embedding methods	Mechanisms, implementation scope and sources
26 attack classes, four codec shortcuts, five pipelines	Transformations and severity controls
25 objective metrics	Definitions, directions and interpretation
Browser research platform	Create, run, inspect and export experiments

Start an experiment

1. Install the package and run the bundled decoding workflow.
2. Start the research UI to define a protocol and select data and conditions.
3. Review the execution plan, run the experiment and inspect failures alongside successful trials.
4. Export the configuration, provenance, result tables and research report.

Scientific interpretation

Each trial embeds a binary payload, optionally applies an attack and extracts with a fresh decoder. Quality scores separate cover-to-stego distortion from stego-to-attacked damage. BER measures payload recovery; a held-out steganalyser estimates detectability. These quantities answer different questions.

Read the protocol, statistical analysis and measurement definitions before comparing methods. Citations identify scientific sources; local adaptations and optional pretrained models are identified in the catalogues. Results reported in publications remain separate from local measurements.

This static site documents the toolkit. The research application runs separately with a Python API, PostgreSQL and the web client. Polski przewodnik · Project and licence.