Google has launched Gemini 3.5 Live Translate, its latest audio model for near real-time speech-to-speech translation across more than 70 languages, aiming to make multilingual conversations feel more natural, continuous, and human-like across Translate, Meet, and developer tools.
TL;DR
- Google’s Gemini 3.5 Live Translate supports over 70 languages, preserves speech tone, pacing and pitch, and works across Google Translate, Google Meet, Gemini Live API, and Google AI Studio.
- Meet will expand from five supported languages to over 2,000 language combinations, while Google is adding SynthID watermarking to generated audio.
Google has revealed Gemini 3.5 Live Translate, a new audio model designed to make live speech translation sound less robotic and more like an actual conversation.
The company said the model can automatically detect over 70 languages and generate smooth, natural-sounding translated speech while preserving the speaker’s intonation, pacing, and pitch. Unlike turn-by-turn translation systems that wait until a person finishes speaking, Gemini 3.5 Live Translate processes speech continuously and stays only a few seconds behind the speaker.
The result, as Google describes it, is a more fluid translation experience without long pauses between speakers.
“Twenty years ago, translation at Google began as one of our pioneering machine learning experiments to turn the science of language into the magic of human connection,” wrote Anuda Weerasinghe, Product Manager, and Tony Lu, Senior Staff Software Engineer at Google. “Today, we’re taking our next step with the release of Gemini 3.5 Live Translate, our latest audio model for live speech-to-speech translation.”
The model is rolling out across several Google products, starting with public preview access for developers through the Gemini Live API and Google AI Studio. Enterprises will get access through a private preview in Google Meet this month, while broader availability is expected later this year. It is also rolling out to users through the Google Translate app on Android and iOS.
For developers, Google says the model can be used to build real-time interpretation tools for multilingual calls, meetings, lessons, broadcasts, customer support interactions, and live events. The company also named Agora, Fishjam, LiveKit, Pipecat, and Vision Agents as developer platforms using the Gemini Live API to support voice translation apps.
Google also said Grab is testing the model to help drivers and travelers communicate during pickups, adding that Grab users make over 10 million voice calls per month through the platform.
One of the most important upgrades is coming to Google Meet.
Google said speech translation in Meet will soon use Gemini 3.5 Live Translate, expanding support from only five languages to more than 70 languages. The update will also enable conversations across more than 2,000 language combinations in one meeting, moving beyond the earlier limitation of translating mainly to and from English.
This could make Meet more useful for global teams, classrooms, international events, and businesses working across multiple regions.
The company is also updating the Meet interface to make speech translation easier to access during meetings. The feature is entering private preview for select Google Workspace business customers this month, followed by a broader rollout later this year.
Google is also bringing Gemini 3.5 Live Translate to Google Translate globally on Android and iOS.
Users can connect any pair of headphones, open the Live Translate feature, and hear translated audio that mirrors the speaker’s tone across more than 70 languages. For Android users, Google is adding a new listening mode that lets people hear translations directly through the phone’s earpiece, similar to a regular phone call.
Google said this could be useful in situations such as guided tours, travel, or quick conversations where users may not have headphones available.
The company added that all audio generated by its models is watermarked with SynthID, an imperceptible watermark woven directly into the audio output to help make AI-generated content detectable and reduce misinformation risks.
Topics for more insights:
SiliconANGLE noted that the model’s continuous stream translation approach makes it different from earlier systems, as it does not need to wait for a speaker to finish before it starts producing translated audio. The publication also highlighted potential use cases across customer support, classrooms, guided tours, ride-sharing, and live broadcasts.
The wider question is whether Google can make this feel reliable in noisy, fast-moving, real-world conversations. If it does, Gemini 3.5 Live Translate could become one of the more practical AI releases in Google’s Gemini lineup, not just for consumers, but also for enterprises and developers building multilingual communication tools.
