False-wake exit cost decides whether people dare to use it

What it is

A character on the TV says the device’s name. The speaker ducks the show and asks what to play. Restoring the room means barging, saying cancel, waiting out a question, bargaining the volume back. False-wake exit cost is the time, turns, and actions needed to push an unsolicited opening back down and return the world to its pre-wake state. Why the hotword false-triggered is an input event. The question here is how hard the dialogue that follows is to tear down. The more expensive the teardown, the less people dare leave always-listen on.

Why it happens

A false wake is an opening with no summons. Once the opening stands, what follows is a real session: it may speak, ask, listen, even act. Getting the prior state back — interrupt the speech, say cancel, wait out timeouts, restore ducked media — sums to exit cost. A non-zero false-open rate times a high exit cost is expected interruption load, set against the value of legitimate wakes. When load beats value, people disable wake, pull the plug, or hide the device. Willingness to use is an expected value on openings, not a grade on skill quality.

Dialogue policy is more controllable here than acoustics. If a false opening rides the same long prompt, the same confirmation, the same timeout as a true session, a true session’s weight is used to punish a mistake. Drive exit cost down: short, barge-able, silence dies it, no execution, ducked media restored at once. Even when acoustic false-positives will not fall, the dialogue side still decides whether the microphone stays.

Studying it

Plant a false opening on playing media. Measure time-to-restore-quiet, turns to cancel, whether ducking occurs, and whether anything executed. Take false-open rate as a given; vary only the dialogue that follows (a long menu versus an earcon that silence can kill). In the field, disable-wake, unplug, and moving the device out of the room are overt willingness measures; pair them with false-open × exit duration. Do not rerun hotword false-accept on this card.

Bystander ratings matter. The person who did not say the wake word still pays the teardown labor; that social bill is charged to the device.

Where it stops holding

Push-to-talk has almost no false openings; this cost does not apply. A false open in headphones has lower social cost and still steals attention; slow exit still hurts. Names that recur in children’s shows are a known trigger; the exit path must stay cheap at that density. A user who meant to wake, treated as a false open and killed in a second, is the opposite harm: exit cost must not be so low that true openings are torn down too. If irreversible actions can fire on a false open, cheap exit will not save it.

Applying it

• Put false openings on a light path: no long prompt, about a second of no-speech then close, no execution on this path, restore ducked media the instant of cancel or timeout.
• Provide a one-word dismiss that works mid-prompt (“cancel,” “stop”), and let barge-in cut an opening that is still being spoken.
• Plant false opens in walkthroughs. Write median time for the room to be quiet into acceptance. If it cannot land around three seconds or less, the always-listen switch will start turning off.
• Count executions after false opens. That count should be zero. Any other number treated an unsolicited opening as consent.

Related

• Same group: M1.12.1 Wake is an explicit handoff of attention · M1.12.2 The opening should hint what this session can do · M1.12.3 Session end needs a perceptible close
• Nearby: C7.08 False Wakes · C7.01 Wake Words · M4.07 Felt Privacy of Always-On Microphones
• Search terms: false-wake exit cost · dismiss cost · unsolicited opening