Proposal: audio-reactive spectral bars for the Pill overlay

[KronosMMXM]

Problem

The Pill currently derives its live shape from a scalar level and decorative motion. Speech with different frequency content can therefore produce similar-looking shapes, and a uniformly active waveform does not clearly distinguish quiet regions.

Proposed change

Use the existing accepted microphone PCM to drive six centered capsule bars by default, with the existing settings supporting 3–8. A bounded visualization queue feeds a serial frequency analyzer; overlapping speech bands and a single attack/release envelope give each bar an independent response. Silence settles to stationary dots. The normal and Spoken Send layouts retain their shell and controls.

The existing transcription/loading compositor shimmer remains unchanged. Recording PCM, microphone capture, recognition, and the scalar meter used elsewhere are preserved.

Implementation and verification

The implementation is submitted in PR #1007, with deterministic PCM fixtures, concurrency/lifecycle coverage, and loading continuity tests. The full Apple Silicon suite passed locally: 1,830 tests, 28 skips, zero failures. Analysis p95 measured 0.23–0.45 ms in Debug. Actual microphone recording and completion were also checked.

I am submitting the implementation for review alongside this Discussion. Maintainer acceptance of the feature direction is pending; this is not a claim that the proposal has already been accepted.

Visual examples

Magnified crops of the actual live renderer using synthetic fixtures, with a composited dark backdrop:

The static images illustrate only the live meter. Loading retains the existing animated sweep.