Google launches Gemini 3.5 Live Translate: a speech-to-speech model for 70+ languages

Google has released Gemini 3.5 Live Translate, a streaming model for simultaneous speech translation in real time that supports more than 70 languages. MarkTechPost reported on June 9, 2026 that the new model translates speech directly into speech, generating audio as a continuous stream and lagging the speaker by only a few seconds. It is available through the Gemini Live API and is built into Google Meet and the Google Translate app.

How streaming translation works

Classic machine translation systems first recognize the whole utterance, then translate the text, and only then synthesize audio. Gemini 3.5 Live Translate works as a stream instead: the model processes incoming sound in chunks and starts generating translated speech almost immediately, without waiting for the speaker to finish a sentence. That is why the delay shrinks to a few seconds, comparable to a professional simultaneous interpreter at a conference rather than the usual pause in translation apps that process a full recording.

Key facts about the launch:

• The model is called Gemini 3.5 Live Translate and belongs to Google's Gemini family.
• It supports speech translation for more than 70 languages.
• It works in streaming speech-to-speech mode, a few seconds behind the original.
• It is available to third-party developers through the Gemini Live API.
• It is integrated into Google Meet and the Google Translate app.

Where the model will appear

Google is building Gemini 3.5 Live Translate into several of its products at once. In Google Meet, the model will be able to translate conversation between video call participants speaking different languages with practically no delay. That closes a long-standing gap in video conferencing: until now, most services relied on translated text subtitles rather than live voice translation. In the Google Translate app, the new model replaces or supplements the previous interpretation mechanisms, making the app's conversation mode more natural for travelers and business negotiations.

The model is also open to developers through the Gemini Live API, an interface for embedding real-time features (audio, video, text) into third-party apps. This means speech-to-speech translation will be available not only in Google's own products, but to any company building call centers, educational platforms, customer support assistants or systems for international events.

A long road to live speech translation

The idea of simultaneous machine translation of speech is not new: translator earbuds and phrasebook-style apps have existed for years. Almost all of them suffered from the same problem, a noticeable pause between a remark and its translation that turned natural dialogue into an exchange of monologues with waiting in between. Each link in the classic pipeline (speech recognition, text translation, voice synthesis) added its own delay and its own source of errors, and mismatches between stages could distort intonation, pauses and even the meaning of what was said.

A streaming architecture that merges all these steps into a single model answers exactly this problem. And the fact that Google is rolling it out straight into mass-market products like Meet and Translate, rather than leaving it as an experimental feature, signals the company's readiness to scale the technology to millions of users.

What this changes for developers and business

A streaming translation model available via API lowers the entry barrier for companies that previously had to either hire a staff of interpreters or integrate separate speech recognition, text translation and voice synthesis services, with a delay at each stage. A single speech-to-speech model simplifies the architecture of such products and potentially reduces overall latency and operating costs.

For end users, the main effect is that machine translation moves closer to the experience of a live simultaneous interpreter in everyday scenarios: video calls, trips, negotiations. Given that Google is developing Gemini as an ecosystem of models while building new capabilities into its mass-market products (Meet, Translate, search and office services), streaming speech translation may fairly quickly become a standard feature rather than a niche capability available only through specialized simultaneous interpretation services.

The question of translation quality in difficult scenarios remains open: overlapping speakers, heavy background noise, or highly specialized terminology, where even professional interpreters make mistakes. The delay of a few seconds that Google claims is a trade-off between speed and accuracy: the less context the model has accumulated before generating a translation, the higher the risk of inaccuracies on long and syntactically complex sentences. Still, the fact that such a feature is now built into everyday tools like Google Meet, and does not require separate specialized hardware, noticeably lowers the barrier to international communication in business and daily life.