@audio/onset

Onset detection — where notes and hits begin. Onset detection functions (ODFs) plus adaptive peak picking.

Two ODFs and a picker. Compute an ODF over the signal, then peak-pick it into onset times. Classical, deterministic, no model weights.

import { spectralFlux, peakPick } from '@audio/onset'

let { odf, hopSize, fs } = spectralFlux(signal, { fs: 48000 })
let onsets = peakPick(odf, { hopSize, fs })   // Float64Array of onset times (seconds)

spectralFlux(data, opts?)

STFT magnitude → sum of positive bin-to-bin differences. Broadband; catches tonal and percussive onsets alike. opts: fs (44100), frameSize (2048, power of 2), hopSize (512). Returns { odf, nFrames, hopSize, frameSize, fs }.

energyFlux(data, opts?)

Per-frame RMS energy → positive first differences. Cheaper, favours percussive/energy onsets (Klapuri, ICMC 1999). Same options and return shape as spectralFlux.

peakPick(odf, opts?)

Adaptive threshold: a frame is an onset if it exceeds delta × the local-mean ODF and is a local maximum. opts: hopSize/fs (time conversion), windowSize (8 frames), delta (1.4). Returns onset times in seconds.

ODF

A symbol key for handing a precomputed ODF result to a downstream tempo/beat pass, avoiding a second STFT.

Notes

FFT via fourier-transform, windows via window-function. Powers @audio/beat's tempo/tracking. MIT.