Google has introduced Gemini 3.5 Live Translate, a new audio model designed to translate speech almost in real time across more than 70 languages. The company says the system is now rolling out in several products, including Google Translate, Google Meet and Google AI Studio, with developer access available through the Gemini Live API.
The launch marks Google’s latest push to make voice translation feel more natural and less delayed. According to the company, the model continuously generates translated speech rather than waiting for a speaker to finish a full turn before responding. That approach is meant to reduce the awkward pauses common in older translation systems while keeping the translated audio close to the original speaker’s timing.
Google says Gemini 3.5 Live Translate is built to preserve key qualities of a person’s voice, including intonation, pacing and pitch. The model also automatically detects supported languages and is designed to handle noisy, unpredictable environments. Google says that could make it useful for live interpretation in meetings, classrooms, broadcasts, multilingual phone calls and other settings where fast conversation matters.
For developers, the model is available in public preview through Google AI Studio and the Gemini Live API. Google is also making it available to enterprise customers in private preview inside Google Meet, starting this month. For general users, the company says the feature is rolling out in the Google Translate app on Android and iOS.
Google says the update to Meet will expand speech translation significantly. The company says the service will support more than 70 languages, up from a previous limit of five, and will enable more than 2,000 language combinations within a meeting. It also says the Meet interface will be updated to make speech translation easier to access.
The new Meet version is being offered first to a limited group of business Google Workspace customers, with a wider rollout planned later this year.
Google Translate users on mobile will also get access to the new system. The company says people using the live translation feature can connect headphones to hear a translation that better mirrors the speaker’s tone across more than 70 languages. On Android, Google is beginning to roll out a listening mode that sends translated speech directly through the phone’s earpiece, allowing users to listen privately without headphones.
Google highlighted potential uses that go beyond one-on-one translation. It pointed to products and partners building on the Live API for live dubbing and simultaneous translation, and said companies such as Agora, Fishjam, LiveKit, Pipecat and Vision Agents are integrating the technology into their platforms.
The company also cited early testing by partners. Grab is experimenting with the model to support near real-time communication between drivers and passengers during pickups. Google said Grab handles more than 10 million voice calls each month. Other partners, including CJ ENM, LiveKit, Vision Agents and Agora, offered positive feedback in comments published by Google, emphasizing translation quality, speed and low latency.
Google said all audio produced by its models is watermarked with SynthID, an imperceptible signal embedded in the output to help identify AI-generated content. The company says the watermarking is intended to support safety and reduce the risk of misinformation as AI-generated voice tools become more widely available.