aTrain is a tool for automatically transcribing speech recordings using state-of-the-art machine learning models, fully offline and without uploading any data. This directory holds the user and reference documentation. For the project overview, badges, and benchmarks, see the main README.
- Installation (end users) — packaged apps (Microsoft Store, Flathub) and installing from source with pip.
- Linux installation — manual command-line setup on Ubuntu / Debian.
- Windows deployment — MSIX rollout on managed machines, for IT departments.
- Linux service — running aTrain as a systemd service, experimental.
- Uninstalling aTrain — removing the app and its data for the MSIX and Flatpak packages.
- Tutorials — importing aTrain output into QDA software (MAXQDA, NVivo on Windows and macOS).
- Security — security assessments, e.g. the OWASP Top 10 for LLM Applications assessment.
- Code signing policy — how release builds are signed, who approves signing requests, and what the app transmits.
- Verifying a release — checksums, Authenticode signature, source tag and SBOM, for packagers and IT departments.
Development setup, the uv workflow, the Docker dev container, building a standalone executable, and the branching/release model live in CONTRIBUTING.md.
- Adding a model — mirroring a model under aTrain-core,
its entry in
models.json, pinned hashes and licence.