Google has deployed Gemini Live Translate across the Gemini Live API, Google Meet, and Google Translate, bringing real-time speech-to-speech conversion in more than 70 languages while preserving each speaker's intonation, pacing, and pitch.
The key engineering difference from earlier turn-based translation systems is continuous streaming: rather than waiting for a speaker to finish a sentence before generating a response, Gemini Live Translate operates in a rolling window that stays a few seconds behind the live speaker. Google says this removes the awkward pauses that have characterized real-time translation tools and produces "fluid audio" throughout a conversation. Audio output is watermarked with SynthID to keep AI-generated speech detectable, per Google DeepMind's announcement of Gemini Live Translate.
The model is launching across three surfaces simultaneously. Developers can access it today in public preview through the Gemini Live API and Google AI Studio. Enterprise customers can apply for a private preview in Google Meet starting this month, where it will expand from the previous five-language limit to all 70-plus supported languages, allowing meetings with more than 2,000 language pair combinations in a single session. Google Translate on both Android and iOS is receiving the model as a global rollout for its existing Live translate feature, with Android users also getting a new "listening mode" that streams translated audio through the phone's earpiece without headphones.
Google's logistics partner Grab is among the early testers, using the API to bridge language gaps between drivers and passengers on its ride-hailing platform, which handles over 10 million voice calls per month. Gemini Live Translate handles multilingual inputs without manual language-selection steps and is built to function in noisy environments. Developer platforms including Agora, LiveKit, and Pipecat have integrated the Gemini Live API to let teams build translation apps on top of the streaming infrastructure without managing the underlying media pipelines.
Twenty years of machine-translation work at Google, which now processes over a trillion translated words per month, underpins the model. The broader Google Meet rollout to all Workspace customers is scheduled for a later date, with no specific timeline announced at launch.
