Agentic data science with the Data Agent Kit
cloud.google.com·6h ago
Data Science
Simon Willison's Blog·27d ago·5 min read
Gemini Live audio
TL;DR
Google has launched Gemini 3.8 Live and 3.8 Live Extended Thinking, two new speech-to-speech models similar to OpenAI's GPT-Live. A web UI was created to interact with these models, allowing real-time voice conversations.
✦ Why It Matters
Explore Gemini Live models today to enhance your voice interaction applications.
Key Takeaways
- Experiment with Gemini 3.8 Live for real-time voice interactions.
- Use the provided web UI to test different voice presets.
- Integrate WebSocket API for seamless audio communication.
- Follow the tutorial to set up your own voice conversation interface.
Full Summary
Gemini 3.8 Live and 3.8 Live Extended Thinking are Google's latest speech-to-speech models, designed to facilitate real-time voice interactions. These models are comparable to OpenAI's GPT-Live family, showcasing advancements in generative language technology.
A web user interface was developed using GPT-6 Astra Extra High, enabling users to select models, voice presets, and input system prompts for voice conversations. The implementation connects directly to a WebSocket endpoint, utilizing the Web Audio API's AudioContext for audio processing.
This setup allows users to interrupt the model while it is speaking, enhancing interactivity. A tutorial is available for developers to get started with the WebSockets API, promoting ease of use and accessibility.