Google DeepMind unveiled Gemini 3.5 Live Translate, a new audio model that performs near real-time speech-to-speech translation across more than 70 languages. The system automatically detects languages and generates natural-sounding translated speech while preserving the speaker's intonation, pacing, and pitch.

Unlike traditional turn-by-turn translation systems, this model generates speech continuously, staying just seconds behind the speaker without awkward pauses. It processes streamed speech to enable seamless multilingual communication, handling unpredictable environments through improved noise robustness.

The rollout begins today across Google products. Developers can access it through the Gemini Live API and Google AI Studio in public preview. Enterprise customers will get private preview access in Google Meet, expanding language support from five to 70+ languages and enabling over 2,000 language combinations per meeting.

Google Translate apps on Android and iOS now support the feature globally. A new listening mode lets Android users hold phones to their ears for private translations. Partners including Grab and CJ ENM are testing the technology, with Grab handling over 10 million monthly voice calls. All generated audio carries SynthID watermarks to detect AI content and prevent misinformation.