Google unveils Gemma Translator offline AI translator for Raspberry Pi 5

Google Creative Lab, together with its Antigravity AI‑agent platform, released an open‑source offline voice‑translation device called the Gemma Translator. The prototype runs on a Raspberry Pi 5 equipped with 8 GB of RAM and includes a microphone, speaker, small display and a rotary‑knob keyboard. Speech is transcribed locally, translated with the lightweight Gemma 4 E2B model via Google’s LiteRT‑LM edge‑inference framework, and synthesized back to audio using the Moonshine model. All processing occurs on‑device, requiring no internet connection and keeping data off external servers.

The hardware design, source code (Apache 2.0 licence) and 3‑D‑printable case files are published on GitHub, showcasing a privacy‑first AI application that can be replicated for multilingual events, editorial workflows or video production. The project highlights how edge AI can reduce reliance on cloud APIs while offering a practical, locally hosted translation tool.