Simon Willison Releases Web UI for Google's Gemini Live Speech API
Developer Simon Willison released an open-source web interface that enables real-time, bidirectional voice conversations with Google's newly released Gemini 3.8 Live speech-to-speech models directly in the browser. The browser-based tool allows users to select voice presets, set system prompts, and interrupt the AI model while it is speaking. By offering a lightweight, library-free client, this tool enables developers to easily experiment with Google's real-time multimodal capabilities without setting up backend servers. It also serves as a practical reference implementation for building low-latency conversational voice agents. The HTML implementation relies purely on native web technology, connecting directly to Google Generative AI's bidirectional WebSocket endpoint (`BidiGenerateContent`) and using the browser Web Audio API's `AudioContext` for audio streaming capture and playback.
## BACKGROUND
Google's Gemini Live API enables low-latency voice interactions by continuously processing audio streams over WebSocket connections rather than traditional text-based request-response HTTP calls. Speech-to-speech AI models understand and produce audio natively, allowing for natural, fluid conversations with immediate feedback and human-like voice nuances.