NativeVoice

Hold a key, speak, let go. The text appears where your cursor is.

NativeVoice is a macOS dictation app for the languages the big tools treat as an afterthought. Dictation on a Mac is excellent in English and the biggest languages, and patchy to absent in the rest: Czech, Polish, Hungarian, Ukrainian and Greek get only basic dictation — no automatic punctuation, no on-device model — and Bulgarian, Estonian, Latvian and Icelandic get none at all, according to Apple's own feature list.

This one treats those languages as the point rather than the long tail. It speaks 36:

• Belarusian, Bosnian, Bulgarian, Catalan, Croatian, Czech
• Danish, Dutch, English, Estonian, Finnish, French
• Galician, German, Greek, Hungarian, Icelandic, Indonesian
• Italian, Japanese, Kannada, Latvian, Macedonian, Malay
• Malayalam, Norwegian, Polish, Portuguese, Romanian, Russian
• Slovak, Spanish, Swedish, Turkish, Ukrainian, Vietnamese

If a language is missing it is because ElevenLabs does not transcribe it yet, not because it was skipped. Anything on this list gets the same behaviour as any other — there is no first-class language here.

An API key of your own

NativeVoice sends your audio to ElevenLabs for transcription and needs your own ElevenLabs API key to work. ElevenLabs bills by usage, so what you dictate is what you pay for — there is no subscription bundled into the app. The key is stored in the macOS Keychain, never anywhere else.

Installation

Requires macOS 13 or later. Universal: Apple silicon and Intel.

1. Download the .dmg from the releases page.

2. Open it and drag NativeVoice into Applications.

3. Open NativeVoice from Applications. It asks for three permissions on first launch, one dialog after another — you do not have to go looking for them:

   • Microphone — to record what you say.
   • Input Monitoring — lets NativeVoice notice when you hold the trigger key. Without it, nothing happens when you hold the key down. macOS hands this one only to a freshly started process, so the app offers to restart itself once you have granted it; let it.
   • Accessibility — lets NativeVoice paste the finished transcript into whatever you were typing into. Without it, NativeVoice still transcribes your speech, but nothing appears where you were typing: it pastes silently into nothing.

   Granting only one of the two is the most common way to end up confused — the app looks like it is working but nothing lands on the page.

4. Enter your ElevenLabs API key when NativeVoice asks for it (also reachable any time from the menu bar icon).

Using it

Hold the right ⌘ key, speak, and release it. NativeVoice transcribes what you said and pastes it wherever your cursor is.

Custom vocabulary

Names, brands, and technical terms that don't exist in a general-purpose language model are the single biggest source of transcription mistakes — fixing this with a vocabulary file helps more than any setting in the app. Edit the list from the menu bar ("Edit vocabulary…"), or directly at:

~/.config/nativevoice/vocabulary.txt

One term per line; lines starting with # are comments.

Building from source

./Scripts/build-app.sh 0.1.0 1

This produces a universal (Intel + Apple Silicon) .app bundle at .build/app/NativeVoice.app. Run ./Scripts/test.sh to run the test suite first.

License

MIT. See LICENSE.