VocaMac

Hold a key, speak, and your words appear wherever you type.
Open-source dictation for macOS. Your voice is transcribed on your Mac.

Download · Website · Models · Discord

Why VocaMac

• Private. Speech is recognized on your Mac. No account, no subscription, no cloud speech service.
• Works everywhere. Text is typed at your cursor in any app: Mail, Slack, VS Code, the terminal, your browser.
• Fast and accurate. Native Apple Silicon engines (Parakeet, Whisper, Apple Speech) on the Neural Engine and GPU.
• Free and open source under AGPL-3.0.

Install

Needs an Apple Silicon Mac on macOS 14 Sonoma or later.

brew tap vocahq/vocamac
brew trust vocahq/vocamac
brew install --cask vocamac

Or download the DMG from the latest release and drag VocaMac to Applications. The app is signed and notarized by Apple.

Nightly builds, building from source, and uninstalling →

Get started

1. Open VocaMac. It lives in the menu bar, not the Dock.

2. Allow the three permissions it asks for:

   Permission	Why
   Microphone	To hear you
   Accessibility	To type the text into other apps
   Input Monitoring	To notice the shortcut in any app (VocaMac asks to reopen once after you allow it)

3. Hold Right Option, speak, and let go. Your words appear at the cursor.

You can also double-tap the key to start and stop, use a separate hands-free shortcut or a mouse button, and press Escape to cancel. Change the shortcut in Settings → Dictation.

The Tiny Whisper model is built in, so you can dictate right away. VocaMac suggests a better model for your Mac, which downloads once.

Features

Dictation

• Type into any app, with live words in a small overlay while you speak.
• Paragraph breaks at long pauses between sentences, and spoken emoji and numbers ("party emoji" → 🎉, "seven thirty pm" → 7:30 pm).
• Personal dictionary for names and jargon ("get hub" → GitHub), plus snippets that expand short phrases. Import or export it as CSV, JSON or a word list, or bring it over from VoiceInk. Say a word to teach VocaMac how your speech model mishears it, and clear out words you haven't said in 60 days.
• Pastes by default, and types the text key by key when an app doesn't take the paste, or always in apps you choose (remote desktops, some games).
• Auto Send: press Return after each dictation, or only when you end with "send it" as its own sentence. Terminals and code editors are left out unless you add them.
• History of past dictations to search, replay, and paste again (⌃⌘V). You can keep chosen apps out of it, or turn it off.

Writing

• Writing styles per app or website: email, chat, notes, code, terminal.
• Cleanup removes "um", false starts, and changes of mind ("let's meet tomorrow, sorry, no, let's meet Friday" → "let's meet Friday", "scratch that") with on-device rules, and a small on-device model tidies the rest.
• Command Mode: select text and say what to change ("make this shorter", "translate to Spanish").

And more

• Transcribe audio and video files, or record system audio.
• A command-line transcriber for scripts.
• Automation with the Shortcuts app and vocamac:// links.
• Installs its own updates and keeps your permissions.

See every feature on vocamac.com.

Privacy

Local speech engines process audio on your Mac. After a model is downloaded, dictation works offline, and nothing about what you say leaves the computer.

Two features can send data elsewhere, and both are off until you set them up:

• Custom Endpoint uploads each recording to a speech server that you host.
• Remote cleanup (Ollama, LM Studio, or an OpenAI-compatible endpoint) sends text to clean up or edit, never audio. API keys stay in your Keychain.

Speech models

Engine	Best for	Runs on
Parakeet (recommended)	Fastest dictation, in English or 25 European languages	Neural Engine
Whisper	The most languages, and translation to English	Neural Engine and GPU
Apple Speech	Built into macOS 26 and later; language packs may download on first use	On device, managed by macOS
Specialized (Moonshine, SenseVoice, GigaAM, Qwen3 ASR, Cohere Transcribe)	Low-memory Macs; Chinese, Japanese, Korean, or Russian; or the highest accuracy in one of Cohere's 14 languages	CPU

Sizes, languages, and the full list are in docs/MODELS.md.

Screenshots

Show screenshots

Dictation	Speech Model
Writing Styles	Dictionary
Smart Cleanup	History
Recording indicator	In-app update

Documentation

Install	Homebrew, DMG, nightly builds, building from source, uninstalling
Speech models	Every engine and model, with sizes and languages
Command-line transcription	Transcribe files from scripts, with JSON output
Troubleshooting	Resetting permissions or settings, update errors
Managed Macs	Pre-approving permissions with MDM
Architecture	How the app is built

The Voca family

VocaMac is the macOS member of Voca.

Platform	Project	Status
macOS	VocaMac	Stable (v1.2.0)
Linux	VocaLinux	Available now (v0.16.0)
Windows	VocaWin	Beta (unsigned v0.1.0-beta.1)
Phone	VocaPhone	Android beta, iOS TestFlight
Gateway	VocaGateway	Early. Optional self-hosted compute, not on-device

Contributing

Bug reports, ideas, and pull requests are welcome. Start with CONTRIBUTING.md, or say hi on Discord.

VocaMac is built on WhisperKit, FluidAudio, sherpa-onnx, and LLM.swift.

License

AGPL-3.0

Star History

Made for the macOS community · @vocahq on X · [email protected]