Phantom Voice
Transcription runs on your Mac using whisper.cpp with Metal acceleration. Your audio and your transcript stay on the machine. Once the speech model is installed, dictation works offline.
Phantom Voice costs $10 once. No account, no subscription, 14-day full trial. macOS 14 or later on Apple silicon.
Download, move to Applications, grant three permissions
From the fn key to text on screen in under a minute
Why microphone, Accessibility and Input Monitoring are each needed
What happens between holding the key and seeing text
whisper.cpp, Metal, and the models you can choose
What stays on your Mac and what touches the network
A second local model that fixes punctuation without rewriting you
The small animated signal that tells you which state you are in
Large v3 Turbo, Medium, Small, Base and what you trade
Hold the key, or press once to start and once to stop
Teach it the technical terms it keeps getting wrong
Editors, terminals, issue trackers, documents, chat
Pick an input device and fix silent recordings
Why local matters#
Most dictation tools send your voice to a server. That is a reasonable trade for a phone note. It is a different trade when you are describing a bug in a private codebase, reading a customer name out loud, or dictating into a terminal connected to production.
Phantom Voice removes the question. The speech model is downloaded once and runs on your Mac. After that, no dictation audio and no transcript leaves the machine.
Where to go next#
New here? Start with Install Phantom Voice, then Your First Dictation.
Deciding whether it fits your Mac? Read Requirements and Pricing and Licenses.