Groq Windows Dictation

A small dictation app for Windows:

• The configured shortcut starts recording.
• Pressing the shortcut again stops recording.
• Audio is sent to Groq Speech-to-Text using whisper-large-v3-turbo.
• The transcript is always copied to your clipboard.
• The text is then pasted automatically into the active window.
• Settings are managed from the system tray, including the Groq API key, microphone, customizable shortcut, and automatic startup.
• You can add names and terms such as Groq and Clinon with the correct spelling to your personal dictionary.
• Explicit word replacements can correct known variants such as Grok or Grog to Groq after transcription.
• The app checks GitHub Releases for updates and can update itself without deleting your API key or settings.
• A small status icon appears centered at the bottom of the screen while the app is in use: recording, transcribing, and then ready for another 3 seconds.

Setup

For development:

.\run.ps1

For normal use, build the Windows app:

.\build-app.ps1

The resulting app is written to dist\GroqInsertDictation.exe.

To install it in your user profile and enable automatic startup:

.\install-app.ps1

This copies the app to %LOCALAPPDATA%\Programs\GroqInsertDictation\GroqInsertDictation.exe.

The Settings window opens the first time you run the app. Enter your Groq API key, optionally select a microphone, configure the shortcut if desired, and leave automatic startup enabled.

To run the tests:

.\bootstrap.ps1 -Profile runtime
.\.venv\Scripts\python.exe -m unittest discover -s tests -v

Settings

Settings are stored in %APPDATA%\GroqInsertDictation\settings.json. The API key is stored in Windows Credential Manager whenever possible.

Open the personal dictionary from Settings to add names, jargon, and terms that are often transcribed with the wrong spelling. The app sends these terms as spelling context in the same Groq transcription request. Because Whisper treats that context as a hint rather than a guarantee, add known mistakes such as Grok → Groq or Grog → Groq in the Replacements section directly below the dictionary. Replacements are applied locally after transcription and before the text is copied or pasted. The existing free-form Prompt field continues to work alongside the dictionary.

The final period is preserved by default. Enable Punt aan het einde verwijderen in Settings if you prefer transcripts without one final period.

• GROQ_MODEL=whisper-large-v3-turbo for maximum speed.
• GROQ_MODEL=whisper-large-v3 for higher accuracy.
• GROQ_LANGUAGE=nl for Dutch; leave it empty to use automatic language detection.
• DICTATION_INPUT_DEVICE=11 to select a specific microphone from the startup list.
• PASTE_AFTER_TRANSCRIPTION=false to copy the transcript to the clipboard without pasting it automatically.

.env remains available as a fallback and migration path, but is no longer required for normal use.

Runtime and build dependencies are kept separately in requirements.txt and requirements-build.txt. The PowerShell scripts reinstall them only when the Python version or dependency files have changed.

Updates

The app checks for a newer GitHub Release when it starts. If an update is available, a window with an update button appears. The updater replaces only the executable; your settings and API key remain in %APPDATA% and Windows Credential Manager.

Note

The app pastes text using Ctrl+V instead of typing it character by character. This is faster and works better with Dutch characters, punctuation, and longer text. Because the transcript is copied to the clipboard first, you can always paste it manually if automatic pasting fails.

Interface and reliability (0.1.19)

The settings window was rebuilt around five pages in a fixed sidebar: Dicteren (shortcut, microphone, behaviour switches), Herkenning (language, multi-line prompt), Woordenboek (words and replacements side by side, no separate dialog), Verbinding (API key with show/hide and a "Verbinding testen" button, model) and Over (version, update check, log file, restart). Unsaved edits are flagged in the footer and confirmed before closing. Launch GroqInsertDictation.exe --settings to open the settings after startup. The UI code lives in settings_ui.py.

The global shortcut no longer uses a low-level keyboard hook (keyboard package). Windows silently removes such a hook when its callback is slow, which is why the shortcut could stop working until the app was restarted. hotkeys.py now registers the shortcut with the Win32 RegisterHotKey API on its own message loop: the OS consumes the key combination, delivers it as a message, and recording starts and stops on a worker thread. Saved shortcut strings such as alt+z, insert or ctrl+shift+f9 keep working. If another program already owns the combination the app starts anyway and asks you to pick a different shortcut.

Clicking the floating status bubble now stops a running recording; when idle it opens the settings.

Geschiedenis keeps the last twenty transcriptions (newest first) with a copy button per entry and a "Geschiedenis wissen" button. It is stored locally in %APPDATA%\GroqInsertDictation\history.json and is also reachable from the tray menu.

The app now ships its own icon (branding.py): embedded in the executable, used by the tray and shown on every window and in the taskbar instead of the default Tk feather.

The Windows widget tests require an interactive desktop. They verify navigation, visible control bounds, dictionary edits, saving, shortcut capture and cancelling without saving.