Gemini 3.5 Live Translate translates speech in real time

Google has launched Gemini 3.5 Live Translate, a model that translates spoken conversations almost in real time while preserving some of the speaker’s tone, rhythm, and intonation. The translated audio arrives with a delay of just a few seconds, without always waiting for the other person to finish speaking.

The difference lies in how it works. Traditional systems usually process an entire sentence and translate it afterward. This model analyzes the audio as it streams and generates the translation continuously, with fewer silences and awkward interruptions.

It automatically detects more than 70 languages and can handle multilingual conversations without requiring you to configure each language manually. It is also designed to work in noisy environments, such as a street, a station, or a call with poor audio quality.

Where it will be available

Google is rolling out the model across several products and services:

• Developers: public preview access through the Gemini Live API and Google AI Studio.
• Businesses: private testing in Google Meet for some Workspace customers starting this month.
• Individual users: global integration in Google Translate for Android and iOS.

In the Translate app, you can connect a pair of headphones and listen to the translation while the other person speaks. On Android, a listening mode is also starting to roll out that sends the translated audio directly to the phone’s earpiece, as if you were answering a call.

For example, you could point your phone at a guided tour in Spanish and listen to an English translation without anyone else hearing the audio. It is designed for quick, discreet situations, especially when you are not wearing headphones.

More languages in Google Meet

Google Meet’s voice translation update will move from a previous limit of five languages to more than 70. It will also support more than 2,000 language combinations in a single meeting, instead of relying only on translations to or from English.

The feature is launching first in a private preview for selected business customers and will roll out more broadly throughout this year. The goal is for meetings with participants from different countries to no longer have to revolve around a single shared language.

What developers will be able to build

The Gemini Live API allows developers to add this translation to calls, classes, broadcasts, and meetings. Companies such as Agora, Fishjam, LiveKit, Pipecat, and Vision Agents already offer tools for managing audio streaming, so developers do not have to build the entire technical infrastructure from scratch.

Grab is testing the system to make communication between drivers and passengers easier during pickups. The platform handles more than 10 million voice calls per month, making this a test of translation in short conversations with background noise.

All audio generated by the model includes an invisible SynthID watermark. This watermark makes it possible to detect that the content was created by AI and is intended to make deceptive use of generated voices more difficult, although it does not by itself eliminate every risk associated with automatic translation.

For you, the most noticeable change will come to Translate: speaking with someone in another language will feel less like exchanging separate phrases and more like having a conversation. Even so, you should watch for accuracy issues with names, accents, noise, and sensitive situations, because reducing the delay does not eliminate translation errors.