Twenty years ago, Translate at Google began as one of the pioneering machine learning experiments to turn the science of language into the magic of relationships. This experiment has been a huge success, with more than 1 trillion words translated every month for billions of users across our products.
Today, we’re taking the next step with the release of Gemini 3.5 Live Translate, our newest audio model for live speech recognition translation.
This model automatically detects over 70 languages and produces smooth, natural-sounding translated audio that preserves the speaker’s intonation, pace, and pitch. Unlike turn-by-turn systems, which wait for the speaker to finish speaking before responding, 3.5 Live Translation continuously generates audio and balances the trade-off between waiting for context to improve quality and immediately translating to stay in sync with the speaker. It provides smooth audio without any awkward pauses and stays behind the speaker for just a few seconds during your session.
Gemini 3.5 Live Translate is rolling out across Google products starting today.
3.5 Build with live translation
Gemini 3.5 Live Translate processes streamed audio and enables more seamless connections between languages. This model handles multilingual input without the need to manually configure settings. At the same time, its noise immunity ensures that applications can cope with noisy and unpredictable environments. Its features can facilitate live interpretation of multilingual calls, meetings, lessons, broadcasts, etc.
