Requirements
| Operating system | macOS 14 or later |
| Hardware | Apple silicon |
| Language | English |
| Default model | Large v3 Turbo, with Base, Small and Medium available |
| Network | Needed once to download the speech model, then optional |
| Permissions | Microphone, Accessibility, Input Monitoring |
| Location on disk | Applications folder, so macOS remembers the permissions |
On Apple silicon specifically#
Transcription runs through whisper.cpp with Metal acceleration. That is what makes local transcription fast enough for a hold-and-release loop rather than something you wait on. Intel Macs are not supported.
On English#
Phantom Voice transcribes English. If you need another language, this is not the tool for you today.
On the network#
The speech model downloads once, on first run. After that dictation works offline. Purchase, model download and update checks still use the internet when needed, and none of them carry your dictation audio. See Privacy on Device.
On disk#
The speech model is the main cost, and its size depends on which one you pick. Optional cleanup adds a second local model of about 1.1 GB, only if you turn it on.
Next#
- Install Phantom Voice — the setup itself
- Choose a Speech Model — the size and accuracy trade
- Back to the documentation home