Google has introduced Gemini 3.5 Live Translate, a real-time speech translation system that the company says works across more than 70 languages while preserving the speaker’s style, intonation and pacing.
The announcement, made on June 9, positions the feature as a step toward removing language barriers in live conversation rather than simply translating text after the fact.
Translation without the pause
Conventional translation tools usually require a speaker to finish a sentence before the system renders it in another language. Gemini 3.5 Live Translate, according to Google, listens continuously and converts speech into the target language almost immediately, attempting to keep the natural flow of human conversation intact.
Google says the system also carries over characteristics of the original voice, including tone and rhythm, so the translated output sounds closer to the person speaking than to a generic synthetic voice.
Where it is available
The feature is rolling out in stages to Google Translate, Google Meet and the Gemini Live API. In Meet, the change expands speech translation from a handful of languages to more than 70, according to reports on the launch. Developers can build the capability into their own applications through the API.
In practice, that means people who speak different languages will be able to hold a conversation and hear a translation with only a short delay, whether in a video meeting or through a third-party app.
DiffusionGemma: faster text generation
Alongside Live Translate, Google introduced DiffusionGemma, an experimental open model. Most language models generate text one token at a time. DiffusionGemma instead produces whole blocks of text at once using a diffusion approach, which Google says can make text generation up to four times faster.
Taken together, the two announcements show Google pairing work on more capable models with technology intended for direct everyday use. Live translation in particular could change how business, education, travel and international communication are conducted, if the system performs in real use as the company describes.
