Google develops AI translator that works without internet

Google has unveiled Gemma Translator, an experimental AI prototype designed to break down language barriers, even without an internet connection. This device processes conversations locally, meaning audio is not sent to the cloud for translation. Instead, the entire process occurs on the device itself.

The system utilizes Google's open-source AI models from the Gemma family, specifically adapted for translation tasks. It currently supports 55 languages. The prototype, built on a Raspberry Pi 5 with 8GB of RAM, includes a touchscreen, microphone, speaker, and language selector. It combines AI models like Gemma 4 E2B, Moonshine, and Kokoro to convert speech to text, translate it, and then play the translated audio.

This offline functionality is particularly beneficial for travelers, those in remote areas, or during flights where internet access is unavailable. It also offers a significant privacy advantage by keeping conversations on the device. Google Creative Lab and Google Antigravity developed the project, which costs an estimated 80 to 100 euros to build and is available under an Apache 2.0 license. While not a commercial product, Gemma Translator showcases the potential for AI models to perform complex tasks on small devices without relying on cloud connectivity.