The Blubber

At Google I/O 2025, Google revealed that it’s introducing real-time speech translation to Google Meet. Powered by a cutting-edge audio language model developed by Google DeepMind, the feature is designed to support smooth, natural conversations between people speaking different languages.

This new capability translates spoken language instantly into the listener’s chosen language, while maintaining the original speaker’s tone, inflection, and emotional expression.

According to Google, the feature opens up a wide range of practical uses. For example, it could help English-speaking children communicate effortlessly with grandparents who speak Spanish. It’s also geared toward multinational businesses, enabling employees from different regions to collaborate and talk in real time without language barriers.

Google claims the system offers very low latency, making it possible for several participants to engage in a real-time multilingual conversation—something the company says was not previously feasible.

When someone speaks, listeners will still hear a soft version of the original voice, layered underneath the translated audio for a more natural feel.

The real-time translation tool will roll out in beta starting Tuesday for consumer AI subscribers. Initially, it will support English and Spanish, with Italian, German, and Portuguese to follow in the coming weeks.

Google also plans to bring the feature to enterprise users, with pilot testing for Google Workspace customers expected to begin later this year.