Dictator Offline Voice to Text

Dictator turns speech into text without sending audio anywhere.

Hold a button on the Dictator keyboard, talk, release — the result lands in whichever app you were typing in. Transcription runs on-device using Parakeet, a fast speech model that fits on Apple Silicon. Optional clean-up uses Apple Intelligence to tidy filler words and add natural punctuation, also on-device.

— Two modes —

Dictate captures fresh speech and pastes it wherever you were typing.

Assist takes the text already in the field plus a spoken instruction and transforms it: shorten, expand, translate, change tone, fix the grammar, turn dot points into a paragraph. The original text returns with one tap of undo.

— Privacy by design —

• Audio never leaves your device. Inference is entirely local. • No account, no sign-in, no analytics. • The model is downloaded once from Hugging Face — that's the only network traffic Dictator initiates. • The keyboard extension communicates with the app via a private App Group on your device.

— Setup —

A four-step welcome guide gets you running: grant microphone access, download the speech model (about 460 MB, one-time), install the Dictator keyboard, enable Allow Full Access. Full Access is required so the keyboard can read the text you're transforming and open the host app to record — both are strictly on-device operations.

— Requirements —

• iPhone with Apple Intelligence (iPhone 15 Pro, iPhone 15 Pro Max, or any iPhone 16) • iOS 26 or later • Around 600 MB of free space for the speech model

— Open source —

Dictator is open source. Source code at github.com/robgough/dictator.

Made by Rob Gough.