Google Unveils Gemma Translator: A Raspberry Pi-Powered Device for Real-Time Offline Translation

On August 6, Google announced the launch of the Gemma Translator, a new device designed to provide real-time, offline translation capabilities. This innovative gadget is powered by Google's advanced on-device AI, enabling seamless communication without the need for an internet connection.

At the heart of the Gemma Translator is the Gemma 4 E2B model. This is an ultra-lightweight and highly efficient open-source generative AI model specifically designed for resource-constrained edge devices such as smartphones, browsers, and single-board computers. The model features a total of 5.1 billion parameters, with 2.3 billion active parameters, ensuring powerful performance in a compact package.

Developed by the Google Creative Lab team, the hardware for the Gemma Translator is built upon a Raspberry Pi 5 single-board computer. The user experience is straightforward: a user speaks into the microphone, and the device translates the speech into the target language, broadcasting the result through a built-in speaker. Simultaneously, it displays both the original and translated text on an attached screen for clarity.

Crucially, the entire translation process is handled locally on the device. Translation inference is managed by Google's LiteRT-LM execution framework, which is optimized for edge devices. For speech-to-text and text-to-speech functions, the device utilizes Moonshine's local AI processing.

The primary advantage of the Gemma Translator is its ability to facilitate direct communication between people speaking different languages, even in remote locations or areas without internet access. By eliminating the reliance on cloud-based services, Google is making real-time translation more accessible and reliable, empowering users to connect with others anywhere in the world.