Local speech recognition
Speech recognition runs entirely on your device using NVIDIA Parakeet. No internet required.
Open source & free forever
Voice to text.
Locally.
Fast, private dictation for macOS and Windows. No cloud, no accounts, no compromises.
Local audio. No account. macOS + Windows.
dybur
Local dictation active
No network activity
Recording
LLM prompt
Review this TypeScript module for race conditions, simplify the state handling, and suggest one focused test that would catch the bug.
Synthetic visualization — no microphone access.
Why dybur?
Dictation without the cloud trade-off.
Built for people who value their privacy and want their tools to just work.
Works where you type
Works with any text field in any application. Just press the hotkey and speak.
No telemetry
No cloud services, no accounts, no telemetry. Your voice data never leaves your device.
Fast insertion
Sub-second transcription latency. Text appears as you finish speaking.
Language detection
Supports 25 European languages with automatic language detection.
Clean output
Automatic punctuation and capitalization. Natural-sounding text output.
Flow
How it works
Get started in seconds. No setup, no configuration, no learning curve.
Default hotkey
Ctrl
+
Shift
+
Space
- 1
Launch dybur
Start the app. It runs quietly in your system tray.
- 2
Focus any text field
Click into any text input in any application.
- 3
Press Ctrl+Shift+Space
The recording pill appears. You're ready to dictate.
- 4
Speak naturally
Talk at your normal pace. Pauses are handled automatically.
- 5
Text appears instantly
Your words are transcribed and inserted in real-time.
Install notes
Requirements
Lightweight and compatible with most modern systems.
Operating System
- macOS 11+ (Apple Silicon)
- Windows 10 / 11
Hardware
- Microphone (built-in or external)
- ~700 MB disk space for speech model
Release builds
Get dybur
Free, open source, forever.
MIT License