Google's Offline Voice Translator Device Hits the Market

The Google-built Gemma Translator is a handheld voice translation device that runs its entire AI processing pipeline locally on a Raspberry Pi 5, without requiring an internet connection. This means that users can speak into it and receive translations in real-time, without relying on cloud infrastructure.

The device's hardware consists of a Raspberry Pi 5 with 8 GB of RAM, paired with a 480×320 touchscreen, a microphone, a speaker, push-to-talk buttons, and a rotary encoder for selecting languages. The entire setup fits inside a 3D-printed enclosure, which is included in the open-source repository.

The software stack on the Gemma Translator uses Google's Gemma 4 E2B model via LiteRT-LM, as well as Moonshine for speech recognition and Kokoro for text-to-speech output. This allows for a complete speech-in, speech-out translation loop to run entirely on the device.

The interface is designed specifically for face-to-face conversations, with two modes that enable either one person or two people to operate their own microphone and language selection independently.