Support voice interaction as another assistant input and output

[asim]

Priority: P1 — bring the requested foreground “Hey Micro” entry point forward alongside direct commands.

Voice should make the same assistant accessible hands-free, without creating another conversation or execution system.

Acceptance criteria

• Add explicit microphone start/stop, recording feedback and transcript review/correction.
• Send voice-derived requests through the same identity, context, command and task paths as text.
• Offer optional spoken responses, mute and interruption controls.
• Handle mobile permission denial, backgrounding, connection loss and accidental recording safely.
• Define recording/transcript retention and deletion clearly; avoid ambient recording by default.
• Test a complete voice request and follow-up on mobile/PWA; measure latency rather than assuming full realtime video infrastructure is required.

Dependencies and boundaries

User-facing acceptance criteria for existing #1487; use its transport work rather than duplicate it. Voice is the first slice; video is not a prerequisite.

Implementation approach

Audit current main first: some foundations and recent fixes already exist. Close remaining user-visible gaps and preserve working behaviour. This issue records the personal-assistant product backlog, not a claim that every underlying capability is absent. Add focused regression coverage for the behaviour changed.

Latest agreed scope: Hey Micro activates existing dictation

Keep the current Speak behaviour. Add opt-in “Hey Micro” detection to activate the existing microphone speech-to-text experience and fill the editable input. The user checks the transcript and presses Send; do not auto-send or change spoken-reply preferences.

This foreground browser implementation is in #1595. Listening must be visible and stop on backgrounding/navigation. Clearly disclose that browser recognition can send audio to its speech provider; do not claim local-only wake detection or OS-level background listening. The wake phrase is activation, not identity or authorization.

A separate speech backend, local wake engine and native/background support are possible later improvements, not prerequisites for this agreed slice. Actual microphone performance on phones remains a device validation item.