On-device models
Models adapted to the glasses' microphones, chipset and thermal limits.
Heads-up transcription and translation for smart glasses: hear what is said around you and see it in your language.
Show live captions of the conversation in front of you.
Read and hear translations heads-up and hands-free.
Keep local notes of meetings and conversations.
Use voice to reach the assistant without reaching for a phone.
Models adapted to the glasses' microphones, chipset and thermal limits.
Low-power capture and local transcription for captions and memory.
Speech and text translation for multilingual conversations.
The glasses capture and stream the audio; the models run on the paired phone. Footprint and power vary by model and configuration.
The glasses stream live audio to the phone for captions and translation. An optional low-power chip can also record speech to flash and sync it in batches. Results come back as captions on the lens or speech through the speakers, depending on the model.
Illustrative architecture. Where processing runs is evaluated per device and configuration.
Start with the phone’s chip, the wireless link, languages and power budget. Then create an application, download the SDK and validate latency, power and memory on your own hardware.
Developer documentation ↗