@audio/mir-tempogram

Tempogram — local tempo over time: an onset envelope (spectral flux) autocorrelated over sliding windows (librosa/MIREX tempogram class).

npm install @audio/mir-tempogram

import tempogram from '@audio/mir-tempogram'

let r = tempogram(data, { fs: 44100 })
// { times: Float32Array (s), bpm: Float32Array (best BPM per window),
//   matrix: Float32Array[] (raw autocorrelation, one row per window, lags minLag..maxLag),
//   odfRate: number (onset-envelope frame rate, Hz), minLag: number, maxLag: number (frames) }

One row of matrix per analysis window; bpm[i] is 60 * odfRate / argmax(matrix[i]). Shorter lags accumulate more autocorrelation terms, so the estimate biases toward the notated tempo over its subdivisions (e.g. picks the quarter-note over the eighth-note pulse).

Options: - fs — sample rate (default 44100, Hz) · frameSize/hopSize — onset-envelope STFT (default 2048/512) · window — autocorrelation window (default 6, seconds) · hop — window hop (default 2, seconds) · minBpm/maxBpm — tempo search range (default 40/240)

Use when: tempo drifts or varies within a track; for a single global tempo estimate use a beat tracker directly (e.g. @audio/beat).

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

Part of @audio/mir.