Nebula: Personal · Helix Nebula · NGC 7293
WhisperFlow — offline voice dictation for Windows
Local, offline, private push-to-talk dictation for Windows 11 (ES/EN): global hotkey, on-device speech recognition, text injected into the focused app.
- Role
- Author
- Period
- Jul 2026 → present
- Instruments
- Python 3.12 · ONNX Runtime · Parakeet TDT · faster-whisper · Win32 API (ctypes) · pytest · ruff
- Repository
- rubo6/whisperflow ↗
★ 0 stars · ⑂ 0 forks · Python · last push Sep 2026
I wanted the feel of cloud dictation tools without sending a single byte of my voice anywhere. WhisperFlow is the result: hold a key, speak, release, and the text lands in the app you were using. Everything happens on the machine: capture at 16 kHz, on-device speech recognition, a small post-processing pipeline, and injection through the clipboard or synthetic keystrokes with a safe preview fallback.
The engineering I am proudest of is the boring part: a written threat model, invariants forced by the config schema (audio is never persisted, dictated text is never logged), a frozen Python runtime so it does not depend on whatever Python the host has, integrity checks on models and binaries, and a test suite around the state machine that coordinates the hotkey thread, the audio callback and the per-dictation worker.
I dictated the answers that shaped this website with it.