Gemini 3.5 Live Translate is Google’s new speech-to-speech translation model, designed to deliver near real-time multilingual conversations across more than 70 languages and integrated into Google Translate, Google Meet and developer platforms.
Google describes Gemini 3.5 Live Translate as an audio model that automatically detects over 70 languages and generates translated speech while preserving a speaker’s intonation, pacing and pitch, aiming to make cross-language conversations sound more natural. Unlike traditional turn-by-turn translation systems that wait for a speaker to finish before responding, the model processes streaming audio and produces continuous translated output, reducing delays that can disrupt dialogue. The company says the system is built to balance the need for contextual understanding with the need to stay synchronized with the speaker, and to handle noisy environments and multilingual conversations without manual language configuration.
The rollout spans several Google products and services. Gemini 3.5 Live Translate is being deployed globally in the Google Translate app on Android and iOS, where it enables live, bidirectional conversation translation and introduces a new “Listening Mode” that plays translated audio through the phone’s earpiece for more private interactions. At the same time, developers can access the model in public preview through the Gemini Live API and Google AI Studio, allowing them to embed low-latency audio translation into communication platforms, mobile applications and other services.
For enterprise users, Google is integrating Gemini 3.5 Live Translate into Google Meet, initially via a private preview for selected Google Workspace business customers. According to the company, speech translation in Meet will expand from a previous limit of five languages to more than 70, supporting over 2,000 language pairs within a single meeting and enabling participants who speak different languages to follow discussions more easily. Google positions this update as part of a broader effort to improve real-time translation and make its collaboration tools more accessible to global teams and organizations.
Under the hood, Gemini 3.5 Live Translate is presented as a low-latency, audio-to-audio model optimized for real-time translation of spoken conversations. Documentation for developers and researchers notes that it can process continuous audio streams and output both audio and text, and that it forms part of the wider Gemini series of multimodal models. The model card also highlights limitations, including occasional inconsistencies in synthesized voices and challenges in language detection for non-native accents or rapid language switching, as well as possible artifacts from background noise.
Google says all audio generated by Gemini 3.5 Live Translate is watermarked using its SynthID technology, an imperceptible signal embedded in the audio to help identify AI-generated content. This measure is positioned as a safeguard against misuse and part of the company’s broader approach to responsible AI deployment, as outlined in its technical documentation.
Read Article: Opendoor winds down India operations, lays off 250 employees and shifts roles to the US
