While speaking, continuous visual feedback must prove the system is capturing

What it is

While talking, people need continuous visual feedback that the system is in capture / listening: not an icon that vanishes after a tap, but a state visible for the whole utterance. It answers “am I still being heard,” and it serves the operator’s sense of progress. A privacy lamp tells the room whether recording is on; this tells the speaker whether this turn is still open.

Why it happens

Voice has no tactile hold of a key. If the interface goes still after onset, people cannot tell listening from already-committed from hung. Continuous feedback makes listening a maintainable state rather than a one-shot acknowledgement. Motion, a highlighted field, or an explicit “listening” label are the state made visible. It must cover onset to endpoint, because any blank in the middle invites a repeated utterance or a look elsewhere. A level waveform is different: it proves sound pressure. Continuous listening state proves the session has not closed, even in a quiet instant.

Studying it

Compare feedback that vanishes after onset, that lasts to the endpoint, and that flickers. Measure repeated onsets, early quits, and subjective “not sure it is listening.” Use gaze to see whether feedback is treated as something to check. Eyes-free conditions should switch to a continuous earcon, or the study is of another channel.

Where it stops holding

In driving and other settings that must not require looking, continuous visual feedback cannot be a required step; use an earcon or a peripheral lamp that does not steal gaze. If a capture indicator can be turned off by the app, continuous feedback is not trustworthy—that is a privacy-layer failure. Very short commands end before feedback is drawn, so this layer barely has time to act.

Applying it

• Keep listening visible from wake or start-listen through the endpoint; do not flash a single frame after onset.
• Give listening a look mutually exclusive with processing and idle, rather than one shared dot for all three.
• Accept with a recording of the whole speaking stretch: feedback must not go blank in the middle.

Related

• Same group: C7.16.2 Level or waveform motion lets people tell the mic is receiving sound, not that the device has failed · C7.16.3 After recognition, mark a shift into processing, distinct from still-listening · C7.16.4 Without visible feedback, silence could be a hang or a normal wait
• Adjacent: C7.07 Privacy Visibility of Voice Input · C7.10 Live Partial Results
• Search: listening state · capture feedback · voice status