Installation and quick start¶

Python package¶

Requires Python 3.10?3.12. The package is distributed on PyPI:

pip install the-a-files

Optional components are provided as extras:

Extra	Command	Enables
`neural`	`pip install "the-a-files[neural]"`	Pretrained neural watermarking baselines (`AudioSealMethod`, `WavMarkMethod`; PyTorch)
`ai`	`pip install "the-a-files[ai]"`	`FgasMethod` and `MosNetMetric` (TensorFlow ≥ 2.15)
`experiments`	`pip install "the-a-files[experiments]"`	Experiment engine (pandas)
`platform`	`pip install "the-a-files[platform]"`	Research platform: REST API with PostgreSQL persistence and the corpus library (FastAPI, SQLAlchemy, Alembic, psycopg)
`dev`	`pip install -e ".[dev]"`	Test and build tooling (pytest, build, twine)

See External dependencies for system-level prerequisites (C++ build tools, FFmpeg).

Usage¶

The bundled evaluation workflow is exposed through the taf-eval entry point, which accepts the name of a packaged scenario or a path to a YAML configuration:

taf-eval direct-no-metrics   # embedding and decoding only
taf-eval full                # embedding, attacks and all metrics

Individual components are available through registry-based factories:

from taf.methods.factory import SteganographyMethodFactory
from taf.models.types import MethodType

method = SteganographyMethodFactory.get(16000, MethodType.LSB_METHOD)
stego = method.encode(cover, message)
decoded = method.decode(stego, len(message))

External dependencies¶

PESQ, when built from source on Windows, may require Microsoft Visual C++ 14.0 or later, available through the Microsoft C++ Build Tools.
FFmpeg must be available on PATH for codec attacks and some format conversions: https://ffmpeg.org/.

Optional models¶

AudioSeal and WavMark load released pretrained weights through their optional packages; first use may require a download. FGAS and MOSNet require the TensorFlow extra.

ViSQOL Audio requires the official Google bindings and libsvm_nu_svr_model.txt, installed separately following the official instructions. The TAF adapter uses audio mode and resamples metric inputs to 48 kHz; this does not restore missing bandwidth. Missing bindings or model files are reported as metric errors. Do not pool audio-mode and speech-mode scores.

For browser-based experiments, follow the UI setup.