klob/research klob.io

Algorithm comparison

Method index

Compare each method by its output, timing requirements, and likely failure modes. Open its chapter for the underlying algorithm and evidence.

OnsetsLive

Envelope threshold

Compare short-window energy with a recent baseline, then debounce attacks.

Very low cost. A sustained pad or a snare roll can swamp a simple threshold.

OnsetsLive-capable

Spectral flux / SuperFlux

Measure positive changes across frequency bands. SuperFlux suppresses motion caused by vibrato.

Effective on attacks. Peak picking that waits for future frames adds delay.

TempoLive-capable

Inter-onset interval voting

Cluster gaps between onsets and vote for a plausible pulse period.

Cheap and interpretable. Subdivisions, syncopation, and missing notes split the votes.

TempoLive-capable

Autocorrelation / tempogram

Score periodicity over many lags and update the scores in local windows.

Reveals multiple tempo candidates. Longer history improves resolution and slows adaptation.

TempoLive-capable

Comb filter bank

Compare resonator responses at candidate beat periods across frequency bands.

Works on streams. Ringing and build-up affect response to a tempo change.

BeatsLive-capable

Grid fit / phase-locked loop (PLL)

Align an oscillator to nearby onset evidence and smooth phase corrections.

Low compute and predictable motion. Can lock halfway between true beats.

BeatsSequence model

BeatRoot agents

Competing agents follow candidate periods and onset alignments.

Can recover from a false start. Published system and live adaptations need separate latency tests.

BeatsOffline

Dynamic programming

Optimize onset evidence and interval consistency over the recording.

Bridges weak evidence. Backtracking uses future material.

BeatsLive

BTrack / real-time PLP

Track or project a local pulse. PLP supplies stability and future beat candidates.

A good interpretable baseline. Sudden stops can invalidate projected events.

DownbeatsLive-capable

Bar-phase templates

Score metrical offsets using bass, timbre, and repeated accents on a known beat grid.

Simple and inspectable. Uniform kicks leave several offsets equally plausible.

PhrasesLive

Bar counter + manual anchor

Count bar groups from an operator-provided musical start.

Predicts ahead. A two-bar extension displaces later cues until re-anchored.

PhrasesOffline

Checkerboard novelty

Look for changes between neighborhoods in a self-similarity matrix.

Finds acoustic boundaries. A symmetric window spends future context.

SectionsOffline

Recurrence / spectral clustering

Combine repeated musical content with local continuity in a graph.

Links returning passages. Cluster resolution changes the structural level.

TempoWindowed

Tempo-CNN

Classify local tempo from an 11.9-second mel-spectrogram excerpt.

Gives rate without a beat offset. Requires a substantial observation window.

DownbeatsOffline

RNN + DBN / madmom

Learn joint beat/downbeat activations, then decode a plausible meter-aware path.

Enforces rhythmic consistency. The classic joint model uses bidirectional evidence.

DownbeatsLive

BeatNet / BeatNet+

Extract causal neural activations. BeatNet uses particle filters for beat phase and bar state.

Supports live tracking. Test recovery on percussion-free passages and blends.

DownbeatsOffline

Beat This!

Use convolutions and attention for beat/downbeat activations with direct decoding.

Strong published F1. Grid continuity can still suffer.

DownbeatsLive

Beat This! Live

Use causal context and shifted targets to anticipate beat/downbeat decisions.

Promising published online scores. Profile full integration and scheduling.

SectionsOffline

All-In-One

Jointly infer beats, downbeats, functional section boundaries, and labels from stems.

Useful for precomputed scene cues. Labels target musical function.

PhrasesOffline

Learned phrase boundaries

Train on independent phrase annotations and decode candidate boundaries.

Closer to melodic phrasing. Scores depend strongly on corpus and matching tolerance.