Google has introduced Gemma Translator, an experimental project demonstrating how artificial intelligence can translate spoken conversations in real time without requiring an internet connection.
The device processes everything locally: it captures speech through a microphone, converts words into text, translates the content, and plays the result using a synthetic voice. Because conversations do not need to be sent to external servers, the system could also offer greater privacy.
The system runs on a Raspberry Pi 5 and uses Google’s open Gemma 4 E2B model along with the company’s LiteRT-LM platform. Google has released the code, documentation, and even the files needed to build the device’s case using 3D printing.
For now, Gemma Translator is not a commercial product but rather an open platform aimed primarily at developers and technology enthusiasts.
Can it work without Wi-Fi or mobile data?
Yes. Once the necessary models are installed during the initial setup, the translator can perform its functions completely offline.

