@doot-games/analysis

Offline spectral analysis of a decoded audio buffer. Pure signal work over raw channel data, so it never touches the Web Audio graph and can run against a synthesized buffer in a test. Depends only on @doot-games/dsp.

API

analyze(buffer, opts?)

Takes a Web Audio AudioBuffer or any object exposing numberOfChannels, length, sampleRate, duration, and getChannelData(c). Returns:

{
  sr, hop, N, frames, duration, binHz,
  flux,           // raw spectral flux per frame
  env,            // normalized onset strength envelope per frame
  bands,          // per frame { lo, mid, hi } energy
  onsets          // peak-picked [{ time, frame, strength, lo, mid, hi }]
}

Options: fftSize (default 1024), hop (default 512), sr target rate for analysis (default 22050).

pickOnsets(analysis, opts?)

Adaptive local-mean peak picking over the envelope. Returns the onset list. Used internally by analyze, exposed for reuse. Options: window, k, delta, minGap.

estimateTempo(analysis, opts?)

Autocorrelation of the onset envelope for BPM and first-beat offset. Returns { bpm, offset, fps } with offset in seconds. A gentle weight toward 128 BPM breaks octave ties. Options: min and max BPM bounds.

Method

FFT per frame, half-wave-rectified spectral flux split by band for the onset envelope, adaptive peak picking for discrete onsets, and autocorrelation of the envelope for tempo. This is the method proven in TREADLE, ported to the shared dsp primitives. An essentia.js WASM backend can be added behind this same interface later as an accuracy upgrade; it is an internal backend, not a user facing engine.