GOOGLE says its AI technologies and products now support more than 300 languages, potentially reaching up to 7 billion people, or about 86% of the world’s population. In an announcement attributed to senior vice-president James Manyika, the company also described a move away from translating speech through an intermediate text transcript. Its Gemini models are being trained to process audio directly, with the aim of retaining tone, emotion and conversational context.
The company announced Gemini 3.5 Live Translate, which supports real-time translation across 70 languages and more than 2,000 language-pair combinations, including code-mixed conversations such as Spanglish and Hinglish. Gemini 3.5 Transcribe is described as Google’s most accurate speech-to-text model, including for noisy environments and technical language. A related Android Gboard feature, called “Rambler”, can remove filler words, correct grammar and punctuation, and apply voice-command edits.
Google also introduced TranslateGemma, a set of open-source models supporting 55 languages and designed to run locally without an internet connection. The company said it is testing a voice assistant called “Ask Viamo Anything” with Viamo in Rwanda, where it has handled more than 2 million voice queries.
In accessibility efforts, Google announced sign-language-to-text support for more than 50 sign languages, although the Pixel 11 launch will initially convert American Sign Language to English through Gboard and Live Transcribe. Google cited local language initiatives including WAXAL, Project Vaani and Language Explorer as part of its wider 1,000 Languages Initiative.