The free, open-source speech-to-text keyboard for iOS.
Self-host Whisper, keep your data, pay nothing.
One command spins up a Whisper server on your machine. Any hardware—Pi, NAS, home server, or cloud VM.
docker compose up -d Install Diction from the App Store. Add it in iOS Settings and point it to your server’s URL.
Settings → Keyboards → Diction Switch to Diction in any app. Tap the mic, speak naturally, and watch your words appear.
Your audio goes to your server. No cloud middleman, no data collection. Fully auditable open-source code.
No subscription, no word limits, no upsells. Self-host on hardware you already own and pay nothing.
Works with faster-whisper, whisper.cpp, OpenAI API, or any compatible endpoint. No vendor lock-in.
Whisper supports 99 languages out of the box. No configuration needed—just speak in your language.
Pure Swift, no third-party SDKs. No tracking, analytics, or telemetry. Every line is inspectable on GitHub.
From tiny (fastest, runs on anything) to large-v3 (most accurate). Pick the model that matches your hardware.
| Diction | Wispr Flow | Apple Dictation | |
|---|---|---|---|
| Price | Free | $15/mo | Free |
| Your data stays | Your server | Their cloud | Apple cloud |
| Open source | Yes | No | No |
| Self-hosted | Yes | No | No |
| Accuracy | Whisper | Proprietary | Poor |
| Languages | 99 | 100+ | ~60 |
Whisper comes in four sizes. Match the model to your hardware.
Fastest. Quick notes, runs on a Raspberry Pi. Sub-second latency.
Best balance. Great accuracy for most uses. Runs on any modern CPU.
Higher accuracy. 4+ GB RAM recommended. 5–10 second latency.
Most accurate. GPU recommended. Best for critical transcription.
$ git clone https://github.com/omachala/diction.git
$ cd diction
$ docker compose up -d
✓ Whisper server running on http://localhost:8000Then open Diction on your iPhone, enter your server’s URL, and start talking. Your audio stays on your network. Always.
Stop paying for proprietary voice-to-text. Self-host Whisper and take back control.
Get started on GitHub