Live translation will now be available on any headphones. Google has updated its Gemini audio models

Google has announced a major update to its Gemini 2.5 Flash Native Audio model. The company is focusing on improving interaction with voice AI agents and is launching a simultaneous translation feature that preserves the speaker’s intonation.

This was reported on Google’s official blog.

What has changed 

Google claims to have improved the model in three key areas to make communication with AI as close to human interaction as possible:

• More accurate function calling. The model has become more reliable when integrated with external tools. It better understands when to bring in third-party data during a conversation and seamlessly weaves it into its response. In the ComplexFuncBench Audio benchmark, the updated Gemini scored 71.5%.
• Following instructions. The developer’s instruction compliance rate increased from 84% to 90%. This means that bots will make fewer mistakes when executing complex scenarios.
• Smooth dialogue. The quality of multi-turn conversations has been improved. AI better remembers the context of previous phrases and does not lose the “thread” of the conversation.

Representatives from Shopify and United Wholesale Mortgage have already tested the update, noting that users often forget they are talking to artificial intelligence.

Live Speech Translation

In addition to improving its business tools, Google has introduced Live Speech Translation for the Google Translate app. This is speech-to-speech technology that allows you to use headphones for simultaneous translation of conversations.

The technology reproduces not only the content, but also the intonation, tempo, and pitch of the speaker’s voice. The system supports over 70 languages, automatically detects the language of the interlocutor, and can process dialogues where several languages are used simultaneously.

The live translation feature is launching today in beta mode in the Google Translate app for Android. In the first phase, users in the US, Mexico, and India will have access. Support for iOS and expansion to other regions are promised to be added in the near future.