BlurtPad: voice notes for Windows

Press Ctrl+Alt+N anywhere, talk, press Ctrl+Alt+N again. The window pops up while it listens, then saves what you said as a note. Open BlurtPad from Start any time to read old notes, newest first. Closing the window keeps it running in the tray (right-click the tray icon: Open, Start with Windows, Quit).

In the window: search your notes (Ctrl+F), click into a note to fix what it heard (it saves on its own), Esc or Discard throws a recording away (Undo brings it back). Settings (the gear, or Ctrl+,): bring the window to the front or stay in the background (a sound and a red dot on the tray icon instead, for games), the shortcut, the microphone, the language, Start with Windows, Check for updates. Team meeting: say how many people and their names, each person reads one sentence so BlurtPad learns their voice, then record the meeting; the note says who said what (click a name to fix it). Voices are remembered for next time. When a new version is out, an Update card appears at the bottom left. Install it from blurtpad.com, or with BlurtPad-Setup-<version>.exe from Releases (no admin needed). Remove it from Settings > Apps like any other app; your notes stay.

Speech is turned into text on your PC by faster-whisper; nothing you say leaves it.

.\.venv\Scripts\pythonw.exe blurtpad.py               # run it
.\.venv\Scripts\python.exe tests\test_notes.py       # checks: note files, updates, the real window, hotkey thread and Whisper

Run from source: Windows 10/11, Python 3.11, then py -3.11 -m venv .venv and .\.venv\Scripts\pip install -r requirements.txt.

• Notes are plain .txt files in Documents\BlurtPad, one per note, named by date and time. Delete in the app sends them to the Recycle Bin. You can edit or delete the files yourself; the app re-reads them when its window gets focus.
• If a recording can't be transcribed, it is kept as a .wav next to the notes instead of being lost (next to the log if the notes folder can't be written; also when Windows shuts down mid-recording).
• Speech model: small.en for English, on the CPU, offline after the first download (about 460 MB). Any other language, or Detect automatically, uses the multilingual small (about 480 MB, downloaded the first time you pick one). model_name() in blurtpad.py.
• Team meetings: Whisper's sentences are matched to the voice-checked people with a speaker model (sherpa-onnx, 3D-Speaker CAM++, about 26 MB, downloaded on first use). Voice fingerprints stay in %LOCALAPPDATA%\BlurtPad\voices.json on your PC; forget them in Settings. Uninstalling deletes them.
• Settings live in %LOCALAPPDATA%\BlurtPad\settings.json (Start with Windows is the registry's Run key). Tests point BLURTPAD_SETTINGS and BLURTPAD_DIR at a temp folder.
• Updates come from this repo's releases (REPO in blurtpad.py). BlurtPad checks soon after it starts, then daily. Update downloads the release's BlurtPad-Setup-<version>.exe, checks it against GitHub's checksum and quits; the installer runs without asking (keeping the old version if it fails), then BlurtPad starts again.
• The installer is per user (no admin), into %LOCALAPPDATA%\Programs\BlurtPad. Installing over a running BlurtPad, or uninstalling, first asks it to quit (BlurtPad.exe --quit, which writes down a recording in progress). Uninstalling also removes Start with Windows, settings.json and the log. Notes and any kept .wav files stay, and so does the speech model in the Hugging Face cache.
• The page is ui/index.html (one file). ui/preview.js documents the page/Python contract and previews the page in a browser: open ui/index.html?preview&state=recording&theme=dark (also &update, &empty, &problem).
• Errors go to %LOCALAPPDATA%\BlurtPad\blurtpad.log.
• The app is about 250 MB (Python, Whisper runtime, WebView2 bridge). The model lives in your Hugging Face cache, not in the app.

License: MIT.