Features
Diction is a voice keyboard for iPhone. One mic button. No QWERTY. Tap, speak, text appears wherever your cursor is. Here is everything it can do.
Context-Aware Text Editing
Diction reads where your cursor is and figures out what you want without you switching modes. Cursor in empty space, it dictates. Cursor in the middle of a sentence, it inserts. Text selected, it rewrites the selection or follows an instruction like "translate to Czech" or "make this shorter". One mic button, the right behaviour every time.
AI Companion
Raw speech is messy. AI Companion removes filler words, fixes grammar, adds punctuation, and produces text that reads like you wrote it. Tuned to sound human rather than machine-cleaned. No em-dashes everywhere, no faintly corporate cadence.
It also takes voice commands. Translate, summarise, rewrite as a list, make it more formal. Anything you can describe in a sentence.
Profile
Tell Diction who you are and how you write. Software engineer with British English. Sales lead who talks to enterprise customers. Doctor who dictates patient notes. A few sentences are plenty. AI Companion uses your Profile on every cleanup, alongside whatever Tone Preset you have set for the app you are in.
Tone Presets
Set a writing style per app. Professional for email, Casual for messages, Friendly for social, or write a custom instruction in plain language. Diction remembers which tone goes with which app and applies it automatically. Same dictation, different output depending on where you are typing.
My Words
Add names, jargon, product terms, and anything else that keeps coming through wrong. Diction uses your custom words to improve both raw transcription and AI Companion accuracy. Add them once in the app, they work everywhere.
Three Ways to Transcribe
Pick the backend that fits how you work. Switch any time in settings.
On-Device . Speech models run locally on your iPhone. No internet required. Audio never leaves your device.
Self-Hosted . Run a Whisper server on your own hardware. One Docker Compose command. Free and unlimited.
Diction One Cloud . Frontier speech models, highest accuracy, zero setup. Includes AI Companion.
Long Dictations
Long messages, long notes, full meetings. Diction handles dictations end to end without truncation. Talk for as long as you need and the full text comes back.
Your Voice Is Never Lost
If transcription fails for any reason, Diction keeps your audio safe and shows a retry button on the keyboard. Network drops mid-recording? Diction falls back to a local model on your iPhone automatically. The result lands exactly where your cursor was, without re-recording a single word.
Transcription History
Every dictation is saved. The history screen splits into recent and full archive. Search highlights matching text so you find what you need fast. Tap any entry to open the full transcript.
If you dictated something ten minutes ago and need it again, you do not have to remember it. Open history, find it, tap, done.
Insights
See how much you are actually dictating. Daily time saved, words per minute, how much faster you are than typing, days used. A stats badge on the home screen shows your usage at a glance.
Preferences
Language, mic timeout (down to "release the moment I'm done"), functional keys, auto start, hand layout, haptics. Every setting in one place.
99 Languages
On-device, self-hosted, and cloud modes all support the full set. Auto-detect is on by default, so you can switch between languages mid-sentence without touching settings.
Works in Every Text Field
Voice keyboards often break in unusual text fields. Terminals, code editors, browser inputs, search bars. Diction is built to work everywhere. If you can place a cursor in it, Diction can insert text into it.
No Word Limits
On-device and self-hosted modes have no word caps, no daily limits, no time restrictions. Talk for five seconds or fifty minutes. There is no meter running. Diction One cloud is unlimited for subscribers.
Privacy
The Diction app contains no analytics, no tracking code, and no telemetry. There is no data collection of any kind in the app.
On-device mode processes audio in memory and discards it immediately. Self-hosted mode sends audio only to your server. Diction One cloud processes and discards. Nothing is stored, nothing is used for training. The server infrastructure is open source.
Your words come through as you said them. No automatic censoring of profanity.