We’re launching Lyria 3.5 in Google Flow Music, with advances across musicality, lyrics, vocals, and creative control
deepmind.google·6d ago
AI Engineering
DeepMind Blog·Dec 12·5 min read
Improved Gemini audio models for powerful voice experiences
TL;DR
Voice interactions often lack clarity and responsiveness, hindering user experience. Google developed the Gemini 2.5 Flash Native Audio model to enhance live voice agents.
✦ Why It Matters
Engineers can utilize the Gemini 2.5 model to create more responsive and accurate voice applications.
Key Takeaways
- Utilize Gemini 2.5 for improved live voice agent capabilities.
- Implement sharper function calling for real-time data integration.
- Enhance customer service with better instruction following.
- Explore live speech translation in the Google Translate app.
How It Works
The Gemini 2.5 Flash Native Audio model improves conversation quality by effectively retrieving context from previous turns, allowing for more cohesive interactions. It also enhances function calling by accurately identifying when to fetch real-time information, seamlessly integrating it into the audio response.