BeautifulVOICE

Speak. It types where you are.

Press a shortcut in any app, say what you mean, press it again — the text lands where your cursor is. Speech is recognized on your computer, never in the cloud.

Free and MIT-licensed · Windows 10/11 and macOS 12+ on Apple Silicon · the speech model (~650 MB) downloads on first start

Three keys, any app

Chats, email, your code editor, a browser form — if it has a text field, you can talk into it.

1

Press the shortcut

A small black pill appears at the bottom of the screen. Its bars follow your voice; a timer counts.

2

Speak naturally

Finished phrases are recognized while you're still talking, so nothing piles up at the end.

3

Press it again

The text is pasted into the field you were typing in — about half a second later on an ordinary CPU.

Pick a model, see how good it is

Every model shows two measures. The built-in voice test runs one recording of your voice through every downloaded model — so you see which one understands you best on your own computer.

Accuracy — 100 − word error rateSpeed — measured on your CPU in the voice test
NVIDIA · Nemotron

Nemotron 3.5 ASR default

0.6B · 651 MB · 32 languages, detects the language itself
Accuracy
91.3%
Speed
fast
NVIDIA · Parakeet

Parakeet TDT v3

0.6B · 670 MB · 25 European languages
Accuracy
94.3%
Speed
fastest
Sber · GigaAM

GigaAM v3

0.2B · 227 MB · Russian
Accuracy
88.8%
Speed
fastest
OpenAI · Whisper

Whisper Large v3 Turbo

0.8B · 1.6 GB · 99 languages
Accuracy
93.0%
Speed
moderate

Published English results from the Open ASR Leaderboard (GigaAM: Russian, its own evaluation). Also in the app: Canary 1B v2, Parakeet TDT v2, Nemotron Speech EN, Whisper Large v3, Distil-Whisper, Whisper Small and Base.

Models page with accuracy and speed meters Voice test comparing models on the same recording

Private by design

Recognition runs on your computer. Audio is never saved or sent anywhere; only the text goes into your history.

Stays out of the way

Lives in the tray, can start with Windows, gives your clipboard back after pasting, keeps dictation out of clipboard history.

Yours to change

Hold-to-talk or press twice, your own shortcut, microphone and history limit. MIT-licensed — fork it, improve it.

The interface speaks nine languages

English中文हिन्दीEspañolFrançaisPortuguêsРусскийDeutsch日本語

Give your hands a break.

Windows: double-click BeautifulVoice-Setup.exe — it installs in a few seconds, no administrator rights needed. If SmartScreen warns about an unsigned app, choose “More info → Run anyway”. A portable .zip is on GitHub.

macOS: open the .dmg and drag Beautiful Voice to Applications. The first time, right-click the app and choose Open, then allow the microphone, and turn Beautiful Voice on in System Settings → Privacy & Security → Accessibility and Input Monitoring so the shortcut and pasting work.