Google Launches Gemini 3.5 Live Translate for 70+ Languages

Key Takeaways

• Gemini 3.5 Live Translate supports 70+ languages with automatic detection and near-instant latency
• The model preserves intonation, pacing, and pitch so translated speech sounds like your voice
• Rolling out to Google Meet, Translate app (Android and iOS), and developer APIs starting this month

Google has released Gemini 3.5 Live Translate, a speech-to-speech AI model that translates conversations in real time across more than 70 languages. Unlike previous translation tools that required specific hardware or turn-by-turn exchanges, this model processes continuous audio while you're still speaking.

The result: conversations that flow naturally, with translated speech following just seconds behind the original. And the output doesn't sound like a robot. Google says the model matches the speaker's intonation, pacing, and pitch.

“We aren't just translating words; we are translating the intent, the pace, and the emotion of the human voice in real-time.”

— Aparna Chennapragada, VP of Product at Google

How It Works

Gemini 3.5 Live Translate is part of the 3.5 model family that Google announced at I/O. Until now, Google had only released the Flash version. A Pro model is expected in the coming weeks.

The translation model handles multilingual inputs automatically. Developers don't need to manually configure language settings. The AI detects what's being spoken and translates accordingly. It also filters out background noise in busy environments.

Google's demos, recorded under controlled conditions, show fluid exchanges where the translated voice maintains the speaker's natural cadence. The company claims the system supports more than 2,000 unique language pair combinations through its automatic detection model.

Where You Can Use It

Gemini 3.5 Live Translate is rolling out across multiple Google products:

• Google Meet: Select enterprise customers get access this month, with a wider rollout planned. Google is updating the Meet interface to make the live translate feature more prominent.
• Google Translate app: Coming to both Android and iOS. You can use any earbuds, or no earbuds at all.
• Developer APIs: Public preview available now through Gemini Live API and AI Studio.

The most notable change for consumers: you no longer need special hardware. Late last year, Google began testing Gemini-based translation with any earbuds, removing the previous requirement for Pixel Buds with an Android phone. This update goes further.

The New Listening Mode

On Android, a new "listening mode" lets you hold your phone up to your ear like a phone call to hear translations. No earbuds required. Google's example: a user listening to a near real-time English translation of a Spanish guided tour directly through their phone's earpiece.

This mode is Android-only for now. iOS users will need earbuds or the phone's speaker.

What Changed from Previous Google Translation

Google has been working on real-time translation for years, calling it one of its "pioneering machine learning experiments." But earlier versions had limitations. You needed specific Google hardware. The translation worked in turn-based segments that broke conversational flow. The output sounded mechanical.

Gemini 3.5 Live Translate shifts to continuous audio streaming. The AI processes and translates while the speaker is still talking, cutting latency and maintaining natural rhythm. The voice synthesis attempts to sound like the original speaker rather than a generic text-to-speech engine.

Developer Access

Developers can start building with the public preview in the Gemini Live API or AI Studio. The model handles language detection and translation automatically, which simplifies integration. Google says developers won't need to manually configure multilingual settings.

The background noise filtering could be useful for applications targeting noisy environments like airports, conferences, or factory floors.

Privacy and Misuse Concerns

Discussion on Hacker News and Reddit has focused on two themes. First, excitement about "universal translator" scenarios for travel and international business. Second, caution about real-time voice synthesis and potential deepfake misuse.

Google is including SynthID watermarking in the generated audio, which embeds imperceptible markers to identify AI-generated speech. Whether this addresses the concerns remains to be seen.

Frequently Asked Questions

Does Gemini 3.5 Live Translate require special earbuds or hardware?

No. You can use any earbuds, or on Android, hold your phone to your ear like a phone call using the new listening mode.

How many languages does Gemini 3.5 Live Translate support?

More than 70 languages with automatic detection, enabling over 2,000 unique language pair combinations.

When will Gemini 3.5 Live Translate be available in Google Meet?

Select enterprise customers get access starting this month, with a wider rollout planned.

Is Gemini 3.5 Live Translate available for developers?

Yes. A public preview is available now through the Gemini Live API and AI Studio.

Does the translated voice sound robotic?

Google says the model matches the speaker's intonation, pacing, and pitch to sound more like the original voice than generic text-to-speech.

Source: Ars Technica