@audio/mir-structure

Structural segmentation — Foote novelty (Foote 2000): MFCC timbre frames → cosine self-similarity → checkerboard-kernel correlation along the diagonal → novelty peaks mark section boundaries (verse/chorus/texture changes).

npm install @audio/mir-structure

import structure from '@audio/mir-structure'

let { boundaries, novelty, times } = structure(data, { fs: 44100 })
// boundaries: number[] (s, detected section edges)
// novelty: Float32Array (one value per frame, checkerboard-kernel score; 0 in the un-scoreable edge margins)
// times: Float32Array (s, frame centers, same length as novelty)

boundaries are peak-picked from novelty: local maxima above mean + sensitivity·std, at least kernel frames apart.

Options: - fs — sample rate (default 44100, Hz) · frameSize/hop — MFCC analysis window/hop (default 2048/1024) · kernel — checkerboard quadrant size, frames (default 16 — wider kernel = broader-scale boundaries) · sensitivity — novelty threshold in std-devs above the mean (default 1, higher → fewer boundaries)

Also exported as an audio.js stat manifest (./audio).

Part of @audio/mir.