Dictate Keyboard
An open-source Android voice keyboard that transcribes on-device or through your chosen provider and connects to self-hosted OpenAI-compatible servers.
Dictate Keyboard is an Android keyboard for people who prefer speaking to typing or find typing uncomfortable. It runs on Android 8.0 and later, with offline speech recognition on the phone and optional cloud or self-hosted providers. The app is open source under Apache 2.0 and builds on FlorisBoard.
Downloaded Whisper, Parakeet and Canary models transcribe without an internet connection. Kroko supports live on-device transcription, so text can appear as you speak. Cloud choices include Soniox, Deepgram, OpenAI and ElevenLabs. For a server you control, it works with OpenAI-compatible backends such as Speaches, faster-whisper-server and whisper.cpp.
You can choose a separate writing model to translate, summarize, shorten or change the tone of dictated text. Custom rewrite prompts handle recurring writing tasks, while vocabulary hints and exact replacement rules help with names and jargon. Rewording requires access to the selected service.
It's also a full keyboard, with glide typing, suggestions, autocorrect and clipboard tools. System voice input and an optional floating microphone let you dictate while keeping another keyboard. It can transcribe shared audio or compatible video files and keep searchable history with configurable audio retention. Wear OS support works through a phone or independently.
Offline audio stays on-device. With your own provider key, audio goes directly to that provider; optional Dictate Cloud routes it through the app's server, which retains neither audio nor transcripts. Only that cloud route requires an account. The app has no ads or telemetry, and general typing isn't recorded or transmitted.
