Google took advantage of Google I/O 2025 to announce the arrival of a new feature within Google Meet. From now on, the videoconferencing service will be able to translate its users' conversations in real time, with "low latency". To achieve this feat, Google Meet obviously relies on Google's artificial intelligence.
The Mountain View giant indicates that the AI model behind the feature is none other than AudioLM, anartificial intelligencecapable ofgenerating audio, such as human voices or music,from a short sound clip. The model is developed by Deepmind, Google's AI subsidiary.
How does Google Meet's real-time translation work?
In concrete terms, the AI will double the words of your interlocutors. This isn't about subtitles, a feature that's already available on the service. Instead of hearing your colleague's voice in their native language, you'll hear the AI's in the language of your choice. In fact, the AI will replace the words of your interlocutor. This will also be entitled to a real-time translation of your responses, which will allow you to communicate without having to learn a new language.
As Google explains, you will first hear the original voice, muted, then the translation will arrive shortly after, which "will cause delays in the conversation". Google specifies that Google Meet is programmed to preserve the characteristics of the user's voice, such as tone, intonation and emotion. Thanks to generative AI, Meet can understand and perceive nuances in people's speech.
A feature reserved for certain subscribers
Initially, Google Meet's real-time translation service is reserved for Google AI Pro and Ultra subscribers, the new $250+ per month subscription that has just arrived in the United States.
Only two languages are supported: English and Spanish. Italian, German, and Portuguese will be rolled out in the "coming weeks". Google plans to make real-time translation available to all companies using Workspace. Google promises "early testing" later this year.
Source: Google
0 Comments