Hands-Free Voice Review (Anki add-on)

Review your Anki decks by voice instead of the keyboard — fully offline.

Speech recognition runs entirely on your computer with Vosk. Nothing is sent to the internet — see docs/PRIVACY.md.

Install

From AnkiWeb (recommended): search for "Hands-Free Voice Review" in Anki (Tools → Add-ons → Get Add-ons) once it is published.

From a file: download the latest .ankiaddon from the Releases page and double-click it, or in Anki use Tools → Add-ons → Install from file.

Restart Anki, open a deck, and study by voice. A microphone is required; allow mic access if your OS prompts. Toggle it under Tools → Voice Control (listening).

The first time you enable it, the ~40 MB speech model downloads automatically (internet required that once); afterwards everything runs fully offline.

How it works

Microphone ─▶ recognizer binary ──stdout──▶ add-on (background reader thread)
               (Vosk, offline)                    │  mw.taskman.run_on_main(...)
                                                   ▼  (UI thread)
                                       state-aware action: reveal / grade / undo

The recognizer is a small standalone program (one per OS) that listens with a restricted 6-word grammar and prints recognized words. The add-on performs the matching action only while you're reviewing, so it can't act at the wrong time.

Configuration

Tools → Add-ons → select this add-on → Config. See addon/config.md. Key settings: spoken words (reveal_words, grade_words, undo_words), response speed (block_seconds), and the noise gate (grade_min_level).

Building from source

The add-on ships a prebuilt recognizer binary per platform. To rebuild them:

• Windows: powershell -File build/build_recognizer.ps1
• macOS / Linux: bash build/build_recognizer.sh
• All three at once: push to GitHub and run the build-recognizers workflow (.github/workflows/build.yml); download the artifacts into addon/recognizer/ as recognizer-windows.exe, recognizer-macos, recognizer-linux.

Get the voice model with build/get_model.ps1 (Windows) or download vosk-model-small-en-us-0.15 from https://alphacephei.com/vosk/models and place it in addon/model/.

Package the shareable file with:

python packaging/make_ankiaddon.py   # -> dist/handsfree_voice_review.ankiaddon

Platform notes

• Windows: works out of the box. PyInstaller .exes are occasionally flagged by antivirus (false positive); allow it if needed.
• macOS: the helper is unsigned, so the first launch may require allowing it in System Settings → Privacy & Security, plus granting microphone access. Signing/notarization (Apple Developer account) would remove the prompt.
• Linux: requires PortAudio at runtime (libportaudio2); it is bundled in the shipped binary but may depend on your audio stack (ALSA/PulseAudio/PipeWire).

License

Add-on code: MIT. Bundled third-party components (Vosk, the Vosk model, sounddevice, PortAudio) keep their own licenses — see addon/THIRD_PARTY_LICENSES.md.

Inspired by williamknows/anki-voice.