VoiceInk transcribes your speech locally on your Windows PC. Hold a hotkey, speak, release — your words appear in any app. No internet, no cloud, no compromises.
Inspired by WisprType • Built for Windows • Open Source (MIT)
From hotkey press to text in your app — the entire pipeline runs on your machine in under a second.
Press and hold your configured hotkey (default: Right Ctrl). Recording starts instantly from your default mic.
Talk in any of 99+ languages. Auto-detection handles the rest. Audio stays in memory — never written to disk.
faster-whisper runs entirely on your GPU or CPU. No data sent anywhere. ~0.3s on CUDA, ~2-5s on CPU.
Transcribed text is injected into whatever app you're using — Word, Slack, VS Code, browsers, anywhere with a cursor.
Every feature is designed for speed, privacy, and reliability. No bloat, no subscriptions, no cloud dependency.
Automatically removes filler words (um, uh, like), fixes self-corrections, adds punctuation, and capitalizes sentences.
Add names, technical terms, medical vocabulary, and brand names. Injected into Whisper's prompt for better accuracy.
System-wide push-to-talk works in any app. Configurable — Right Ctrl, Right Alt, CapsLock, F13, or custom combos.
Text appears in any app with a text cursor — Word, Outlook, VS Code, Slack, Discord, Chrome, Notion, and more.
Lives quietly in your taskbar. Icon changes color to show state — idle (gray), recording (red), processing (yellow), done (green).
Every transcription saved locally in SQLite. Search, copy, delete, or export as TXT/CSV. Never leaves your device.
Works in any app with a text cursor
VoiceInk is built with a zero-trust network model. It literally doesn't need the internet after the initial model download. No telemetry, no analytics, no accounts, no cloud — just you and your PC.
After model download, works completely offline. No API calls, no web requests.
Zero analytics, zero tracking, zero phone-home. We don't even collect crash reports.
Audio is processed in-memory and discarded. Never written to disk, never uploaded.
Transcription history stored in local SQLite. Never synced, never uploaded, fully yours.
No sign-up, no login, no email. Download and run. That's it.
Every line of code is auditable. Verify our claims yourself. Fork it, modify it, share it.
Five model sizes for every use case. Smaller models are faster; larger models are more accurate. All run locally.
Quick notes
General use
Best balance
Professional use
Maximum accuracy
NVIDIA CUDA
~0.3s for 10s audio • float16 • Requires CUDA toolkit
CPU Fallback
~2-5s for 10s audio • int8 • Works on any PC
WisprType brought local voice-to-text to macOS. VoiceInk brings it to Windows — and makes it open source.
| Feature | WisprType macOS | VoiceInk Windows |
|---|---|---|
| Platform | macOS (Apple Silicon) | Windows 10/11 |
| Local Processing | ||
| Smart Typing | ||
| Custom Dictionary | ||
| Open Source | ||
| GPU Support | Apple Neural Engine | NVIDIA CUDA + CPU |
| Cost | Free | Free (MIT License) |
| Windows Support | ||
| Linux Support | Planned | |
| Languages | 99+ | 99+ |
Clean, auditable Python architecture. Each module has one job. Easy to extend, easy to audit.
Records 16kHz mono PCM via sounddevice. In-memory numpy array. No disk writes.
faster-whisper wrapper. CUDA/CPU auto-detect. Custom vocab via initial_prompt.
Regex-based filler removal + punctuation. Optional LLM post-processing.
Win32 SendInput or clipboard+paste. Restores original clipboard content.
Global push-to-talk via keyboard library. Non-blocking background thread.
JSON-based vocab manager. Categories: Names, Technical, Medical, Legal, Brands.
pystray system tray with state icons. Menu for quick toggles and settings.
SQLite local storage. Search, copy, delete, export TXT/CSV. Never uploaded.
A phased approach to building a production-ready, distributable Windows app.
Working proof-of-concept with core pipeline.
Polished enough for daily personal use.
Complete graphical interface for all settings.
Packaged installer for non-technical users.
Download VoiceInk, pick your Whisper model, and start talking. No account needed. No internet required after setup. Just your voice, your machine, your text.
Or build from source:
$ git clone https://github.com/voiceink/voiceink
$ cd voiceink && pip install -r requirements.txt
$ python main.py