Google has unveiled a translation model capable of real-time interpretation in more than 70 languages.
On Tuesday, Google announced it would roll out "Gemini 3.5 Live Translate" — a translation model built on its latest AI — across the Google Translate app.
Google said the new model moves away from consecutive interpretation toward what it calls "continuous real-time generation," an approach closer to simultaneous interpreting. The model dynamically balances translating speech instantly against waiting for additional context to improve accuracy. As a result, the interpreted audio lags the original speaker by only a few seconds.
Unlike the previous version, which required users to select a language in advance, the new model automatically detects the language being spoken and translates accordingly. It can also handle multilingual conversations in which multiple languages are mixed.
Google has also improved the underlying voice synthesis technology so that the translated audio preserves the original speaker's accent, speaking style and pitch, delivering a more natural-sounding result. Previously, users had to plug in earphones to use voice interpretation on the Google Translate app for iPhone and Android. Now they can simply hold their smartphone to their ear — as they would during a phone call — and hear the translated audio without earphones.
Google said it plans to bring the same feature to Google Meet, its video conferencing platform, to support multilingual meetings and similar use cases.
Major service companies that previewed the model, including Grab and Agora, praised its capabilities. Baek Hyun-jung, chief AI officer at CJ ENM, who took part in early testing, said the model "demonstrates quality impressive enough to deliver a more vivid experience to audiences in Korea and around the world."
yckim6452@heraldcorp.com
