Google integrates neural network into Translate: speech now translated without pauses

Google has announced a significant update to its translation service. The new audio model Gemini 3.5 Live Translate enables real-time speech translation with virtually no lag and is already available to users.

Key features

The model supports over 70 languages, including Russian. Unlike previous systems that waited for the end of a phrase before beginning translation, Gemini 3.5 Live Translate works continuously. It preserves natural intonation, speech tempo, and even individual voice characteristics of the speaker, making the translation more natural and comfortable to follow.

Users can now:

• Listen to translations of lectures, tours, and presentations almost synchronously with the original
• Use "listening mode" in the mobile Google Translate app, where the translation is output through the smartphone speaker

Expansion to other services

The update also affects the corporate service Google Meet. The system can now simultaneously translate speech from multiple participants speaking different languages. Previously, Meet's capabilities were mostly limited to English-to-English and back translation. In the initial phase, the new feature will become available to select categories of Google Workspace subscribers.

For developers

Google has opened access to the model via the Gemini Live API and Google AI Studio platform. The tool is currently in preview testing mode but is already available for integration into third-party applications.

In brief

Google has released Gemini 3.5 Live Translate — a powerful audio model that translates speech in real time without noticeable pauses, preserving intonation and voice characteristics. The new technology is already integrated into Google Translate on Android and iOS, works in Google Meet, and is available to developers. Over 70 languages are supported. This is one of the most significant updates to the service in recent years, greatly enhancing ease of use.