@audio/defeedback-tracker

Howl trajectory tracking — persistence + stability + narrowness gating; the music discriminator

import tracker from '@audio/defeedback-tracker'

Stateful factory that turns per-frame @audio/defeedback-analyzer candidates into confirmed howls. A candidate must clear the pnpr/phpr narrowness gates, persist at a stable frequency (within freqTol) for minFrames consecutive updates, and not be decaying (trend >= growth) before it's reported — this is what keeps sustained musical tones from being mistaken for feedback.

let t = tracker({ minFrames: 3, pnpr: 18, phpr: 15 })

let howls = t.update(candidates)   // candidates from defeedback-analyzer, one call per hop
// → [{ freq, level }, ...] confirmed, growing or stable howls

t.reset()                            // clear all tracked trajectories

Returns { update(candidates): Array<{freq, level}>, reset(): void }. A track is dropped after being unseen for more than 2 updates.

Use when: deciding when to deploy a notch — feed confirmed howls to @audio/defeedback-notchbank.deploy(freq).
Not for: spectral peak detection itself — use defeedback-analyzer.

Part of @audio/defeedback — the defeedback family umbrella.

MIT © audiojs