@audio/mir

Music information retrieval — chroma, chord, key, and MIREX-parity analysis atoms.

Atoms

| Package | Status | What | |---|---|---| | @audio/mir-chroma | ✔ | 12-bin pitch-class profile — PCP and NNLS methods | | @audio/mir-chord | ✔ | Chord detection — 24 binary templates + Viterbi smoothing | | @audio/mir-key | ✔ | Key detection — Krumhansl-Schmuckler profiles | | @audio/mir-tempogram | ✔ | Local tempo over time — autocorrelated onset envelope | | @audio/mir-tonnetz | ✔ | Tonal centroid features — chroma projected onto fifths/thirds circles | | @audio/mir-structure | ✔ | Structural segmentation — Foote novelty on MFCC self-similarity | | @audio/mir-downbeat | ✔ | Downbeat estimation | | @audio/mir-melody | ✔ | Continuous melody F0 contour — YIN per frame | | @audio/mir-multif0 | ✔ | Multiple simultaneous F0 — Klapuri iterative spectral subtraction | | @audio/mir-fingerprint | ✔ | Audio fingerprinting — Shazam-class landmark hashing | | @audio/mir-similarity | ✔ | Audio similarity metric | | @audio/mir-transcribe | ✔ | Polyphonic transcription | | @audio/mir-drums | ✔ | Drum transcription | | @audio/mir-coversong | ✔ | Cover song identification |

ML-tier tasks (genre, mood, tags, stem separation) are deferred — they need hosted model weights, which conflicts with the no-ML-in-the-hot-path stance.

chroma/chord/key originated in pitch and were extracted here.

Usage

import { chroma, chord, key } from '@audio/mir'

let c = chroma(samples, { fs: 44100 })         // Float32Array[12]
let ch = chord(c)                              // 'C', 'Am', …
let k = key(c)                                 // { key: 'C major', scores: [...] }