Voice AI today answers everyone in the room. Attention Labs is the engagement-control layer that decides which speech is actually addressed to a device or agent, in real time and before it acts, so AI responds only when it should & stays silent on the rest.

Our first system, Selective Auditory Attention (SAA), tags every utterance as device-directed, human-directed, or ambient, holds anything below the engagement threshold, and forwards only the addressed audio to any downstream model, on-device or in the cloud, fail-closed by design. It's the missing layer between the microphone and the model: addressee routing, engagement gating, and a clean audio handoff that drops into any voice stack, whatever AI sits behind it.

We're building toward a world where people and AI share a conversation the way people always have: listening, taking turns, and matching the rhythm of the group. Our team, NeurIPS (2023) and CES (2026) award winners spanning data science, neuroscience, and ML engineering, is backed by Techstars, Andreessen Horowitz, and Motivate, with published research behind the approach.

attention labs

Org chart

Teams

Offices