Imagine a world where language barriers simply melt away, allowing conversations to flow effortlessly, regardless of the tongues being spoken. For two decades, Google has been on a pioneering journey to turn the complex science of language into the magic of human connection. What started as an ambitious machine learning experiment now translates over a trillion words every month for billions of users worldwide.
Today, we’re thrilled to announce a significant leap forward in this mission with the release of Gemini 3.5 Live Translate. This groundbreaking audio model delivers near real-time speech-to-speech translation, empowering truly seamless communication across more than 70 languages.
The Breakthrough: How Live Translate Works Its Magic
At the heart of Gemini 3.5 Live Translate is its ingenious ability to process speech continuously. Unlike older “turn-by-turn” systems that force awkward pauses while waiting for a speaker to finish, our new model translates on the fly. It masterfully balances the need for immediate translation with gathering enough context to ensure high-quality output, staying just a few seconds behind the speaker.
This innovative approach results in smooth, natural-sounding translated speech that beautifully preserves the speaker’s original intonation, pacing, and pitch. The model automatically detects over 70 languages, handles multilingual inputs without manual configuration, and boasts remarkable noise robustness. This means crystal-clear communication, even in loud or unpredictable environments, making it ideal for live interpretation in calls, meetings, lessons, and broadcasts.
Empowering Developers and Global Connections
The power of Gemini 3.5 Live Translate is accessible through the Gemini Live API, making it incredibly easy for developers to integrate sophisticated voice translation capabilities into their own applications. Platforms like Agora, Fishjam, LiveKit, Pipecat, and Vision Agents are already leveraging this API. They handle the complex real-time media streaming infrastructure, freeing developers to focus entirely on crafting exceptional user experiences.
Real-world testing is already underway, demonstrating the model’s transformative potential. For instance, our partners at Grab are trialing Gemini 3.5 Live Translate to facilitate near real-time multilingual communication between drivers and travelers during pickups. Considering Grab users make over 10 million voice calls per month, this integration could revolutionize interactions and significantly improve their service.
The feedback has been overwhelmingly positive, with companies like CJ ENM and LiveKit praising 3.5 Live Translate’s impressive translation quality, accuracy, and remarkably low latency. This is truly a game-changer for anyone looking to build robust, real-time voice translation solutions.
Bringing Live Translate to Your Everyday Google Experience
Soon, the advanced capabilities of 3.5 Live Translate will enhance speech translation within Google Meet. This update, launching in private preview for select Google Workspace customers this month with a broader rollout later this year, promises to significantly upgrade the meeting experience by:
- Providing more natural and fluid translations.
- Delivering translations faster than ever before.
- Performing better in noisy meeting environments.
- Enabling continuous translation, eliminating disruptive pauses.
Beyond professional settings, Gemini 3.5 Live Translate is also rolling out globally on the Google Translate app for both Android and iOS. Simply connect any pair of headphones, and you’ll experience a seamless, tonally accurate translation that mirrors the speaker’s nuance across 70+ languages, bringing effortless global communication right to your pocket.
For Android users, an innovative new ‘listening mode’ is also becoming available. This feature allows you to hear translations directly through your phone’s earpiece, just like a regular call. It’s perfect for situations where you need quick, discreet translations without headphones, such as hearing a Spanish tour guide translated into English, straight to your ear.
In our commitment to responsible AI development, all audio generated by our models, including Gemini 3.5 Live Translate, is imperceptibly watermarked with SynthID. This ensures that AI-generated content remains detectable, helping to prevent the spread of misinformation and upholding the integrity of communication. This thoughtful approach underscores our dedication to safety and responsibility, as detailed in our model card.
Source: Google DeepMind Blog
