What you need
- A Mac with Apple silicon (M1 or newer).
- macOS 14 or later.
- About 8 GB of free disk space.
- 16 GB of RAM recommended.
- Internet only the first time: the models (about 3.6 GB) download once.
DictDact
Free voice dictation for Mac. It runs 100% on your Mac and learns your voice.
Hold the Ctrl key and speak or whisper. It can be Spanish, English or both in the same sentence. Let go, and clean, structured text is pasted where your cursor is.
How to download and install it
Four steps. The second one is only needed the first time you open it.
Download the .dmg file
Go to the downloads page on GitHub and get DictDact-0.3.0.dmg. Open it and drag DictDact into the Applications folder.
- First time only
Open it and click “Open Anyway”
The free build is not notarized by Apple yet, so the first time macOS says it cannot verify the developer. Go to System Settings → Privacy & Security, scroll down and click “Open Anyway”. You only do this once.
Let it download the models and turn on the permissions
The Welcome window downloads the models with a progress bar and walks you through the permissions: Microphone, Accessibility and, on macOS 26, Input Monitoring. Turn on DictDact in each list.
Hold Ctrl and talk
Let go of the key and the text appears where your cursor is, in any app.
Keyboard shortcuts
- Hold Ctrl
- Dictate. Let go and the text is pasted.
- Ctrl twice
- Hands-free lock: dictate without holding the key.
- Ctrl + Alt + H
- Opens the DictDact window.
- Ctrl + Alt + V
- Pastes the last dictation again.
What it does
Dictate in any app, in Spanish, English or both
It works anywhere you can type: emails, notes, chats or documents. You can mix Spanish and English in the same sentence.
If you whisper, DictDact boosts your voice by up to +40 dB so it is still understood. Before pasting, a local language model adds punctuation and cleans up the text.
It transcribes with Whisper large-v3-turbo or Parakeet on Apple MLX, and uses a local Qwen model for punctuation and cleanup. Everything runs on your Mac.
It learns how your names are spelled
Speech models often misspell names. In the DictDact dictionary you save how each word is written, and DictDact recognizes sound-alike forms: if the model hears “Hirox” or “Hierox”, it pastes “Hyrox”.
It also suggests words mined from your own dictations. When you correct a dictation with “Fix → Save & learn”, it remembers it for next time. And it reads the names visible on your screen.
No dictation gets lost
Every dictation is saved to the history before it is pasted. If it landed in the wrong place, Ctrl + Alt + V pastes it again.
You can search your dictations and see how many words you dictate per day.
Voice bank: always learning your voice
It is optional. If you turn it on, every dictation is also kept as studio-quality audio (24-bit FLAC). DictDact grades the room noise and files your recordings by language: your Spanish on one side and your English on the other.
That way, what you already dictate every day becomes the material to clone your voice, with your own accent and pronunciation. 30 usable minutes per language is enough for an ElevenLabs Professional Voice Clone; 2 to 3 hours is best. You can also make the clone locally.
To be clear: DictDact does not fine-tune the speech model on your voice. The voice bank collects your voice for a clone, and the dictionary learns your words.
Screenshots with sample data.
Everything stays on your Mac
Once the models are downloaded, DictDact works without internet. Your voice, your dictations, your dictionary and your voice bank are stored on your Mac.
Nothing leaves your Mac unless you export it yourself. The code is open: anyone can review it on GitHub.
Frequently asked questions
- Is DictDact free?
- Yes. It is free and open source, under the MIT license. You download it from the project page on GitHub.
- Does it work on Windows or on an Intel Mac?
- No. For now it only works on Macs with Apple silicon (M1 or newer) running macOS 14 or later.
- Does my voice go to the cloud?
- No. Transcription and text cleanup run on your Mac. It only uses the internet the first time, to download the models. Nothing leaves your Mac unless you export it yourself, for example if you upload your voice bank to ElevenLabs.
- How does it learn my voice?
- In two ways. The dictionary learns your words: the names you save, the suggestions from your dictations and your corrections. The voice bank, if you turn it on, keeps your dictations as audio so you can clone your voice with ElevenLabs or locally. DictDact does not fine-tune the speech model on your voice.
- Why does macOS say it cannot verify the developer?
- Because the free build is not notarized by Apple yet. Go to System Settings → Privacy & Security and click “Open Anyway”. You only do it once.
Download it for free
Mac with Apple silicon and macOS 14 or later.
Download free for MacCode, downloads and release notes: github.com/abrahamolverad/dictdact · MIT license