Google unveiled Gemini 3.5 Live Translate, a speech-to-speech model that captures a speaker's tone, pacing, and pitch while translating across more than 70 languages. The system adds SynthID watermarks for security and aims to sound natural rather than robotic. This builds on Google's years of real-time translation work, previously limited to specific hardware.

The model processes speech continuously with minimal latency, keeping just seconds behind natural conversation flow. Unlike earlier versions that required Google phones or earbuds, this release expands availability across the Google ecosystem. Developers can access a public preview through the Gemini Live API or AI Studio, with automatic language detection and multilingual input handling.

Background noise filtering works in busy environments, making the technology practical for real-world use. Google previously launched only the Flash version of Gemini 3.5, with a Pro model expected soon. The company has demonstrated real-time translation capabilities at events for years, gradually bringing the feature to broader audiences through the Translate app.

Gemini 3.5 Live Translate represents Google's push to make multilingual communication seamless. By preserving vocal characteristics and working across its ecosystem, the tool could reshape how people interact across language barriers in meetings, travel, and daily conversations. The technology demonstrates measurable progress toward truly instantaneous translation.