In a June 10 announcement about the Gemini 3.5 Live Translate model, Google stated that it is introducing translation features to the Meet and Translate applications in several markets, including Vietnam.
Previously, these applications could translate Vietnamese but displayed subtitles on the screen. The new feature, however, allows spoken translation to overlay the original audio through headphones, with low latency, similar to cabin translation. For example, in a conversation with an English speaker, users can open the app, select the English-Vietnamese language pair, and a Vietnamese voice will play simultaneously with the original audio through the headphones.
Users in Vietnam can utilize this feature via the Google Translate app on iOS and Android devices by selecting Live Translation in the bottom left corner of the screen. A necessary condition is to connect to headphones to hear the translation. Google stated that the app can accurately reflect the speaker’s tone in over 70 languages.
For Android users, Google has begun rolling out a mode that allows them to listen to the live translation through the speaker. “Just hold the phone to your ear as if making a regular call, and the translated audio will be transmitted directly to you,” the announcement stated. This feature is useful in situations where users want to quickly hear the translation without others hearing it or when they do not have headphones available.
The Google Translate app interface on June 10, featuring the Live Translation function. Photo: Lưu Quý
The feature also works in the Meet application, using the Gemini 3.5 Live Translate model, supporting over 70 languages instead of just 5 as before, including Vietnamese. It allows for back-and-forth conversations between over 2,000 language pair combinations in a meeting. The Preview version of this feature has just been rolled out to select enterprise customers of Google Workspace before expanding later this year.
Gemini 3.5 Live Translate is Google’s latest audio model, providing near-instantaneous spoken translation capabilities. The company noted that the model can automatically recognize languages and create a translation voice that is “smooth, natural, while maintaining the speaker’s tone, speech rhythm, and pitch.”
To achieve this, Google does not translate sentence by sentence as traditionally done, but continuously generates spoken translations, balancing the wait for additional context to improve translation quality and immediate translation to maintain synchronization with the speaker. The company mentioned that the translation could be “a few seconds delayed” during conversations.
In addition to the Meet and Translate applications, Google is also rolling out the feature for developers through the Gemini Live API and Google AI Studio, enabling the development of voice translation applications. Grab is one of the partners testing this model to facilitate real-time multilingual communication between drivers and passengers at pickup points. Currently, 10 million voice calls through the Grab platform are conducted in this manner each month, according to Google.
“During the testing of Gemini 3.5 Live Translate, we appreciate the model’s ability to automatically recognize multiple languages and accurately translate speech with low latency,” said Philipp Kandal, Product Director at Grab.
To prevent voice translation from being misused to create misinformation, the developer stated that all audio will be watermarked with SynthID, helping to identify AI-generated content.
