Free & open source · macOS

Speak.
It’s typed.

VoiceYak is push-to-talk voice to text for your Mac. Hold a key, say what you’re thinking, and release. It’s pasted wherever your cursor is. The AI dictation runs on your machine, so nothing you say ever leaves it.

Free forever · Apple Silicon · macOS 26.2+

Hold the key below. Watch this line.

0:00 Transcribing

Click and hold the key, or hold space.

0:00 → 0:03

From held key
to pasted text.

Hold

Keep a key down. Fn, Right Option, whichever you pick. It never types a stray character.

Speak

The pill listens while you talk. Say a sentence or dictate for minutes. It keeps up.

Release

Let go. Transcribed on-device, pasted right where you were typing. In any app.

VoiceYak's Voice Model pane showing NVIDIA Parakeet TDT 0.6B running on-device

On-device AI

A frontier model,
living on your Mac.

NVIDIA’s Parakeet TDT, the same family that tops open speech leaderboards, quantized to int8 and running locally on Apple Silicon. Accuracy you’d expect from the cloud, latency you wouldn’t.

VoiceYak's custom dictionary editor

Custom dictionary

It learns
your words.

Product names, jargon, the way you actually talk. Fix a misheard term once and VoiceYak gets it right from then on. Your dictionary is yours, stored on disk, editable any time.

VoiceYak dashboard with dictation stats

Quiet by design

A menu bar app,
not a fleet of windows.

VoiceYak lives in your menu bar and stays out of the way. A dashboard when you want it, a floating pill while you speak, and nothing at all the rest of the time.

Privacy

Nothing leaves
this machine.

No cloud transcription. No account. No telemetry. The only network requests VoiceYak ever makes are a one-time model download and a release check you explicitly opt into. Don’t take our word for it: the source is public.

No cloud No account No telemetry No subscription MIT licensed

42 seconds

Watch it work.

Requires
macOS 26.2+, Apple Silicon
Model
NVIDIA Parakeet TDT 0.6B, int8 (~460 MB)
Engine
sherpa-onnx speech to text, fully offline
Languages
English, or the 25-language multilingual model
License
MIT, free forever
Install
Drag-and-drop DMG, ~2 minutes to first dictation

Stop typing.
Start talking.

Download VoiceYak

github.com/g-ghuman/VoiceYak