Breaking

Google Opens Gemini 3.5 Live Translate API for Real-Time Voice Translation in 70+ Languages

June 12, 2026 at 17:37 EDT

Foundation Models
Media Generation
Software Dev & Coding

Google has unveiled Gemini 3.5 Live Translate, its latest audio model that enables low-latency, near real-time voice translation across more than 70 languages, now available to developers in public preview through the Gemini Live API. A demonstration showed the model ingesting a live broadcast and translating it continuously with synchronized transcripts, letting developers add near real-time voice translation to their own apps.

June 9, 2026 · Google

Gemini 3.5 Live Translate: one model for near real-time voice translation

Now in public preview via the Gemini Live API and AI Studio, it replaces the old multi-stage pipeline with a single end-to-end audio-in, audio-out model — low latency, synced transcripts, and a speaker's own intonation preserved.

70+

languages supported

~2,000

language pairs

end-to-end model, audio in → audio out

From a chained pipeline to one model

Fewer stages means less latency and less context lost in hand-offs.

Traditional — 3 separate stages

Speech → Text

→

Machine Translation

→

Text → Speech

Gemini 3.5 Live Translate — single model

Audio in

→

End-to-end S2ST

→

Translated audio + synced transcript

What it does in a single session

Auto language detection

Multilingual input without manual switching

Native audio output

Preserves intonation, pacing and pitch

Noise robustness

Strong filtering across noisy scenes

Synced transcripts

Text aligned to the translated audio

Model ID gemini-3.5-live-translate-preview · access via Python google-genai & WebSockets · partners LiveKit, Agora, Pipecat

Early praise

Partners "blown away" by speed, accuracy and liveliness
Called a SOTA result — low latency, high accuracy
Single-API design preferred over stitching models

Stated limitations

Voice can drift after pauses or with multiple speakers
Detection struggles with accents, similar or fast-switched languages
Noise filtering strong but not perfect; artifacts possible

Consumer rollout underway: integrated into Google Translate on Android/iOS and a private preview of Google Meet for select Workspace enterprises, with a broader Meet rollout planned later in 2026. No public pricing identified.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.