Speak a phrase, see it translated, hear
it back — with the network switched off. No accounts, no servers, no API keys.
Your voice never leaves the device.
Linguago runs Google's Gemma 4 model locally. It transcribes your speech and
translates it in a single pass, so there is no cloud round-trip and nothing to
intercept.
How it works
Records 16 kHz audio and hands it straight to an on-device model
Returns the transcription and translation together
Speaks the result using your phone's own voice engine
Deletes the recording as soon as it has been read
What it needs
Requirement
Detail
Device memory
6 GB or more
Storage
About 3 GB for the model
First launch
One 2.6 GB download, then fully offline
In beta.
Speech recognition can mishear you, and translation quality varies by
language. If a result looks wrong, type the phrase instead — the text box
gives more reliable results than the microphone.