Google Meet to Feature Real-Time Speech Translation Powered by DeepMind

Google has unveiled a groundbreaking feature for Google Meet at Google I/O 2025: real-time speech translation. This innovative capability leverages a large language audio model developed by Google DeepMind, promising to facilitate seamless and natural conversations between individuals speaking different languages.

The speech translation feature in Meet is designed to translate spoken words into the listener’s preferred language in real time, all while preserving the speaker’s voice, tone, and expression. This nuanced approach aims to maintain the integrity and emotional context of the original message.

According to Google, this feature has a wide array of potential applications. From enabling English-speaking grandchildren to communicate effortlessly with their Spanish-speaking grandparents, to fostering real-time collaboration among global colleagues in multinational companies, the possibilities are vast.

Google emphasizes the low latency of the speech translation, which allows multiple participants to engage in fluid conversation—something the company claims was previously unattainable. The system is designed to overlay the translated speech onto the original voice, allowing listeners to faintly hear the original speaker’s intonation.

The real-time speech translation feature is set to roll out in beta to consumer AI subscribers starting this Tuesday. Initially, it will support English and Spanish, with plans to add Italian, German, and Portuguese in the coming weeks.

Google is also actively developing the speech translation capabilities for business applications within Meet, with early testing slated for Workspace customers later this year. This move indicates Google’s commitment to enhancing global communication and collaboration in professional settings.