Gemini Live audio

Google has released two new conversational AI models focused on speech: Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking. These systems provide continuous voice‑to‑voice interaction, offering an experience comparable to OpenAI’s GPT‑Live family. To make testing easier, the author employed an advanced language model (GPT‑6 Astra Extra High) that produced a simple web interface, free of external dependencies, allowing users to select the model and voice preset, optionally enter a system prompt, and hold a real‑time conversation directly from the browser. The tool uses a WebSocket supplied by the Gemini API and the Web Audio API to capture and play audio, enabling the model to be interrupted while it is responding. The source code is available on GitHub, and Gemini Live’s official tutorial explains how to initiate the WebSocket connection.

This news matters because it expands the options for developers and users looking to integrate AI‑powered spoken interactions, reducing reliance on third‑party libraries and simplifying the deployment of voice‑assistant or customer‑service applications. Moreover, the availability of an “extended thinking” model hints at improvements in real‑time reasoning and coherence of generated responses.

Read the original article on Simon Willison

This summary is an informational synthesis produced by dataqbs.com. All rights to the original content belong to its author and the cited media outlet. We act solely as curators of technology news and claim no authorship.