Wake is an explicit handoff of attention

What it is

Finishing a wake word is not merely flipping a microphone from off to on. It is an attention handoff: the user has paused other talk, oriented to the device, and offered the floor. The system’s first job is to catch that handoff — earcon, light, a particle “mm” — so both sides know the session is open. How a hotword is spotted is an input event. What matters here is the dialogue act after the spot: whether the summons was answered.

Why it happens

Human openings often run a summons–answer: a name or a “hey,” met by gaze, a particle, or “yeah” as a display of availability. The wake word is the summons; the earcon or first packet of feedback is the answer. Until the answer, the user does not know they hold the floor. They repeat the wake, or pour the command into a session that is still closed. A command poured into a closed state sounds like recognition failure; the root is an unfinished handoff.

Handoff fails in two directions. The summons goes out and the answer is late or missing: the user talks into an empty floor. The answer sounds and there was no legitimate summons (a false wake): that error’s cost sits on the way out. Push-to-talk makes the summons a press; the handoff completes on the finger, and the spoken-opening sequence no longer applies. Treat wake as a pure switch and you miss its timing as the first dialogue move: the answer must land before the user starts the business, or the business falls into the seam.

Studying it

Measure time from wake-word offset to a system availability signal (earcon, light, particle), and the rate of repeat-wakes inside two seconds. When the signal is late or absent, repeat-wakes should rise — that is the behavioral mark of an unanswered summons. Use conversation analysis to see whether device openings form a complete summons–answer adjacency pair. Do not use hotword detection rate.

Controls can be light-only, earcon-only, both, and a deliberate delay of three to eight hundred milliseconds. Also score whether the user starts the command before the signal (talking into the seam). Do not run keyword false-accept curves on this card.

Where it stops holding

Push-to-talk has no spoken summons; the press is the handoff. Demanding an extra “I’m listening” then gets in the way. Products that claim “just speak, no wake” make the handoff implicit; failure looks like “I spoke and it didn’t take,” not “I called and it didn’t answer.” In a multi-party room, who was summoned is unsettled: the answer should tell the people not addressed that the floor was not handed to them. If only hearing or only vision is available, the answer has to land on the channel they can use.

Applying it

• Every successful wake must produce a perceptible availability signal before the user starts the business utterance. A long brand sting can wait; the answer itself must be short and immediate.
• Write “talking into the seam” as a fault: availability later than command onset. Fix timing before blaming recognition.
• Pick a primary channel among light, earcon, and spoken answer, and keep a substitute for eyes-free or ears-free situations.
• Film the mouth: the command should start after the answer. If most people start before the earcon, the handoff window is too slow for that group.

Related

• Same group: M1.12.2 The opening should hint what this session can do · M1.12.3 Session end needs a perceptible close · M1.12.4 False-wake exit cost decides whether people dare to use it
• Nearby: C7.01 Wake Words · C7.08 False Wakes · M1.08 Turns and Floor Management
• Search terms: wake as attention handoff · summons-answer · engagement