Google DeepMind brings sign-language AI to Pixel users

Google DeepMind has introduced SL2T, a massively multilingual sign-language-to-text model that is moving into consumer products. The first rollout powers ASL-to-English sign dictation in Gboard and Live Transcribe on Pixel 11, letting users sign into places where they would normally type, including search, messages, documents and Gemini prompts.

The model is trained on more than 100,000 hours of data across more than 50 sign languages, with roughly a quarter of the data in ASL. Google DeepMind says sign languages require true translation rather than word-by-word transcription because they have their own grammar and rely on simultaneous movement of hands, arms, torso, head and face.

For privacy, SL2T does not translate the raw camera feed directly. An on-device MediaPipe model converts the signer into pose landmarks, and the server-side model translates those geometric coordinates into text. Google says more devices and more sign languages will follow.

The launch matters because accessibility AI often stays trapped in demos. This is a concrete product path for Deaf and hard of hearing users, and it gives sign language processing a real consumer surface instead of treating it as a lab-only computer-vision benchmark.