NativeVoice

Dictation in your language.

If you dictate in Czech, Polish, Ukrainian, Greek or thirty-two other languages, macOS has never really had you covered. NativeVoice is built the other way round: hold a key, speak, let go, and the text appears where your cursor is.

Download for macOS Source

Free · open source · pennies per use · macOS 13+ · Apple silicon and Intel

🇧🇾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

All thirty-six behave the same. There is no first-class language here.

One menu, one gesture

Hold right ⌘, speak, let go. There is no window to learn — everything the app can do lives in a single menu, and this is all of it:

Trigger key
Right Command, or Shift together with it.
Language
Follows the system until you pick one.
Max recording length
Stops a recording you forgot about, so it cannot run up a bill.
History
Off by default. On, it keeps the last few transcripts so you can put one back on the clipboard.
Play sounds
A note when recording starts and when the text lands.
Paste automatically
On. Off, the text waits on the clipboard for you.
Remove filler words
Off by default, because it rewrites what you said.
Edit vocabulary
Your own names and jargon, one per line.
Open log
What the app did and why, in plain sentences.
Launch at login
Off by default.
Check for updates
Manual only. Nothing is downloaded unless you ask.

Setting it up

Ten minutes. You bring your own ElevenLabs key: you pay them directly for what you dictate and nothing goes through anyone else. Their free tier is 10,000 credits a month and transcription costs 330 credits a minute — about half an hour of dictation every month for nothing. Check their current pricing before you rely on that.

  1. Drag NativeVoice into Applications and open it. It has no window; look for it in the menu bar.

  2. Make a key at ElevenLabs → Settings → API Keys and copy it at once — they show it only the first time.

  3. Menu → Settings… → paste → Save. It goes into the macOS keychain and the app never shows it again.

  4. Allow Input Monitoring in System Settings, then quit and reopen the app. macOS gives the permission only to a fresh process, so nothing changes until you do.

  5. Allow Accessibility too. This one fails quietly: without it your words are transcribed and the paste is thrown away, so you hear the finishing sound and nothing arrives.

Two things that fix most mistakes

Turn the microphone up

A quiet signal changes digits before it garbles words, which is the dangerous way round. Measured here: at a −41 dB peak “one hundred and forty-two” came back as “one hundred and sixty-two”; at −26 dB it was right. Speech should peak near −12 dB, and the app tells you when yours does not.

Add your own words

Names and jargon are what transcription gets wrong, and they are a short list only you know. One term per line, read before every transcription, so edits apply immediately.

~/.config/nativevoice/vocabulary.txt

Build it yourself

No dependencies, nothing hidden, about 3,500 lines.

git clone https://github.com/trustbe/nativevoice.git
cd nativevoice
./Scripts/build-app.sh 0.1.0 1