Krux
March 28, 2026
Google's New Voice Model Works in Noisy Rooms
Published: March 28, 2026 at 12:38 AM
Updated: March 28, 2026 at 12:38 AM
100-word summary
Google released Gemini 3.1 Flash Live, a real-time audio model built to handle voice conversations even when background noise would normally trip up AI assistants. The model improved its score on a complex audio benchmark to 90.8%, meaning it can actually complete multi-step tasks while you're talking. It's rolling out in 200+ countries through Gemini Live and a new Search Live feature that lets you troubleshoot problems by voice instead of typing. Developers get API access, and enterprises can plug it into customer service tools. All output gets watermarked with SynthID so you can tell when audio came from AI. Voice assistants might finally work in the real world, not...
What happened
Google released Gemini 3.1 Flash Live, a real-time audio model built to handle voice conversations even when background noise would normally trip up AI assistants. The model improved its score on a complex audio benchmark to 90.8%, meaning it can actually complete multi-step tasks while you're talking. It's rolling out in 200+ countries through Gemini Live and a new Search Live feature that lets you troubleshoot problems by voice instead of typing. Developers get API access, and enterprises can plug it into customer service tools. All output gets watermarked with SynthID so you can tell when audio came from AI.
Why it matters
Voice assistants might finally work in the real world, not just quiet demo rooms.