Wake-Word-Free Addressee Detection for Wearables.

Addressee detection determines, in real time, whether speech at the on-body microphone is directed at the device or at another person in the environment. SAA makes this decision in real time, before ASR runs, and forwards only addressed audio to your stack. A wake word only detects a trigger phrase; it cannot determine whether the surrounding utterance was actually intended for the device. Addressee detection closes the gap that wake words leave open, and fails closed when uncertain.

A wake word tells you that a specific phoneme sequence was spoken, not whether the utterance was actually directed at the device. In always-on wearable environments, a wearer quoting the trigger phrase to a colleague, reading aloud, or standing in a room where someone else says the phrase all register as activations. Wake words also require the wearer to prepend a phrase to every natural command, which breaks down in the professional and social settings where wearables are being deployed. SAA operates on the audio at the on-body microphone, working after a wake word fires or instead of one, and fails closed when uncertain.

SAA runs as a hosted cloud service. Audio from the on-body microphone streams to the hosted service at server.attentionlabs.ai, and the addressee decision returns in real time through a thin client SDK. No model runs on the wearable hardware by default. Embedded deployment is available as a separate enterprise license for programs with strict power, privacy, or IP control requirements, and is scoped on a call after an initial evaluation.

The addressee decision returns in real time from the hosted service, before your ASR would have begun processing. No model runs on the wearable hardware in the default hosted configuration. The device-side cost is limited to the audio buffer handoff to the client SDK; no inference runs on the device. For programs where the hosted round-trip is unsuitable due to intermittent connectivity or power constraints, embedded deployment is available as a separate enterprise license.

SAA sits between the on-body microphone and your ASR. Client SDKs in JavaScript/TypeScript and Python stream audio from your existing pipeline to server.attentionlabs.ai and return the addressee decision. The integration surface is a single audio buffer handoff and a binary or probability output. No change to your model, ASR, or UX layer is required.

SAA is hosted and fails closed. Audio that is not addressed to the device is not forwarded to your ASR or model; it stops at the gate. This narrows the category of ambient capture that GDPR data minimization principles and HIPAA minimum-necessary rules require you to justify, which is a common blocker in enterprise wearable procurement. A Data Processing Agreement is available for regulated deployments.

Embedded deployment is available as a separate enterprise license for programs that require it, for example where strict privacy architecture, intermittent connectivity, or power constraints make a hosted round-trip unsuitable. It is not the default public motion and is scoped on a call after an initial cloud evaluation.