Google's Gemini 3.5 Live Translate enables real-time multilingual speech in 70+ languages

Google has released Gemini 3.5 Live Translate, its latest audio model aimed at supporting live speech-to-speech translation. The model automatically detects more than 70 languages and generates fluent, natural-sounding translated speech, preserving the original speaker’s intonation, pacing, and pitch.

Unlike traditional turn-by-turn systems, Gemini 3.5 Live Translate produces translated speech continuously and stays only a few seconds behind the original speaker. This approach reduces awkward pauses and creates a more natural flow in conversation.

Expanding on its capabilities, the model processes spoken language as it streams and does not require manual configuration for multilingual inputs. Its noise robustness enhances performance even in loud or unpredictable environments. These features enable live interpretation in various scenarios, such as multilingual calls, meetings, educational settings, and live broadcasts.

Today’s launch brings Gemini 3.5 Live Translate to multiple Google products. Developers can access a public preview via the Gemini Live API and Google AI Studio. Enterprises gain private preview access this month through Google Meet, and the feature is available to everyone in Google Translate on Android and iOS devices. To support transparency and help prevent misinformation, all audio generated with this model is imperceptibly watermarked using SynthID technology.