Showing the keyboard should pause recognition so key-clicks are not taken as speech

What it is

Once a soft or hardware keyboard is the current input source, recognition should pause automatically (and preferably stop capture). Clicks, tap noise, and muttering enter the microphone and decode as garbage inserted into the document. The pause is channel exclusion, not throwing the whole voice session away.

Why it happens

Keyboard activity is a strong acoustic event: mechanical switches, screen taps, even a fingernail on capacitive glass sit in the speech band. If ASR is still streaming, those pulses become insertions, or they tangle with the characters being typed and “recognized” fake words. Focus in a text field plus a visible keyboard is a strong enough signal: the effector is now a finger. Pause can mean stop recognition, close the mic, or mark audio as non-command. The session need not end: resume when the keyboard hides or the user taps the mic again. The typical accident without a pause is a sudden run of characters in the document unrelated to the keys. This is not bystander attribution: the noise source is the user’s own hand.

Studying it

Type while recognition is still on; count noise-words inserted and whether typing is rewritten by a speech hypothesis. Compare “pause on keyboard shown,” manual mic-off, and no pause. Test a mechanical keyboard and a touchscreen. Dependent measures include false insertions and whether people notice.

Where it stops holding

Typing on an external keyboard beside a far-field speaker while also talking to the speaker is intentional parallelism; pause policy is per device: a phone soft keyboard should pause, a desktop command speaker need not. Voice-controlling a keyboard (“next key”) is voice as a keyboard substitute; pausing ASR contradicts it. Whisper is so quiet that key-clicks are relatively louder, which makes pausing more necessary.

Applying it

• On soft-keyboard show, pause recognition and stop the waveform; do not auto-resume capture on hide until the user taps the microphone.
• During the pause, switch to a non-listening state so it does not look as if recording continues.
• Accept by summoning the keyboard mid-dictation and typing hard; the document must not gain recognition garbage in time with the keystrokes.

Related

• Same group: C7.17.1 Switching from voice to keyboard should keep the caret and existing text, not reset the edit · C7.17.3 Undo in mixed input must cover content from both sources · C7.17.4 Voice-to-keyboard is usually for fine correction; keep that path short
• Adjacent: C7.16 Visible Feedback for Voice Input · C7.05 Noisy Environments
• Search: key click · pause recognition · keyboard shown