An Android keyboard that actually listens offline

What it does

Transcribro is an Android keyboard and system speech-to-text service that transcribes voice input using the OpenAI Whisper family of models running locally via whisper.cpp. It plugs into any app that accepts voice input, either as a dedicated keyboard or as the device’s default speech recognition provider. Silero VAD handles voice activity detection so it knows when you’re actually talking.

The interesting bit

Most speech keyboards are thin clients for cloud APIs; this one pulls the entire Whisper inference pipeline onto your phone. The README steers users toward Accrescent because it considers that channel more secure than GitHub Releases, and it publishes the APK signing certificate for manual verification—which tells you exactly who the project is for.

Key highlights

• Runs Whisper inference entirely on-device through whisper.cpp, keeping audio local
• Doubles as a system-wide speech-to-text service for other apps, not just a keyboard
• Uses Silero VAD to detect when you’re speaking
• Available via Accrescent (recommended by the developer) and GitHub Releases
• Currently English-only; broader language support is planned

Caveats

• Only English is supported right now; other languages are on the roadmap but not yet available
• The project is upfront about this limitation, so polyglots will need to wait for issue #18 to resolve

Verdict

Anyone who wants to dictate passwords, notes, or search queries without trusting a cloud provider should look here. If you need multilingual transcription today, this isn’t your keyboard yet.