Google has finally revealed the technical details of their real-time speech-to-speech translation (S2ST) system, a technology already powering the Pixel 10 and Google Meet. The announcement, made on November 20, 2025, breaks down how Google achieves near-instant translation using the speaker’s own voice.
Unlike traditional cascaded systems, Google’s end-to-end streaming model can translate speech with only about a two-second delay – a major leap in natural, real-time communication.
This new publication is not a product launch but an explanation of the architecture that has been quietly been running inside Google Meet and Pixel 10 devices for several months.
At Google I/O 2025, held from May 20-21, 2025, the company unveiled real-time English to Spanish speech-to-speech translation in Google Meet. However, no technical details were offered then. The demonstration video used during the conference is the same one featured in this latest release, confirming that the system has been active behind the scenes for months. The feature initially rolled out in beta to Google AI Pro subscribers.
Read also: Terrorists Attacks School In Niger State, Kidnaps Students And Staff
In September, Google expanded Google Meet’s AI live speech translation to four additional languages — Italian, German, French, and Portuguese — marking the system’s shift “from AI research to reality.”
Throughout the summer, Google also revealed that Pixel 10 phones are capable of real-time translation during phone calls, preserving the voice and intonation of each speaker. Those technical specifics were withheld at the time as well.
Google Connects the Dots
With this week’s research post, Google has officially confirmed that the S2ST systems used in Google Meet and the Pixel 10 rely on the same training data and architecture.
“The new end-to-end S2ST technology has been launched in two key areas,” Google said, adding that “it is now available in Google Meet servers and as an on-device feature embedded in the new Pixel 10.”
