Short version: your voice never leaves your phone.
OpenEar is a dictation keyboard. Speech is recorded only when you tap the microphone and release/stop, and transcription runs on the device itself.
- Records audio only while you are actively dictating. There is no background recording, no always-on microphone service, and no hidden capture.
- Runs speech recognition locally using a Whisper model you explicitly download once. Dictation works offline afterwards.
- Inserts recognitions directly into the text field you had focus in. If the field changes before insertion finishes, nothing is inserted.
- Never uploads audio, transcripts, or text from your keyboard sessions.
- Never includes analytics, advertising, tracking SDKs, crash-upload, or any account system.
- Never writes transcript text to the clipboard automatically.
- Never saves a transcript unless you explicitly enable optional local history (off by default, stored only in this app's private storage, excluded from backups).
- Never requests contacts, SMS, location, storage, or accessibility permissions. It cannot enable itself as your default keyboard by itself; Android always requires you to do this.
| Permission | Why |
|---|---|
RECORD_AUDIO |
Dictation. Requested the first time you tap the microphone. The microphone is released as soon as you cancel, stop, or dismiss the keyboard. |
INTERNET |
Model download only. The transcription engine has zero network I/O; removing this permission would only break new model downloads. |
- Models live in private app storage; you can delete them any time and reclaim the space.
- Optional history is local-only, deletable per-entry, and clearable.
- "Reset settings" restores defaults; "Clear history" and "Delete downloaded models" do exactly what they say.
OpenEar handles your spoken content on this device, but once text is sent, WhatsApp/Telegram/email and their servers normally get it — OpenEar cannot change how those apps handle your data. Audio or text processed by OpenEar itself is never transmitted, and release builds keep diagnostic logs free of audio and transcript contents.
Because OpenEar uses whisper.cpp offline, accuracy varies by device, accent, noise, and background apps. Very quiet speech and heavy code-switching can still produce errors; you can edit or re-record rather than accept a guess.