Blog · Explainers

Beat detection: how software finds the beat in a song

Beat detection analyses the audio for sudden jumps in energy — onsets — then finds the repeating interval that best explains them. The result is a list of timestamps in seconds, which an editor turns into cut points. It is accurate on percussive music and least accurate on tracks with no clear drum.

It is worth knowing roughly how this works, because it explains the two cases where it gets things wrong — and both have easy fixes.

Step one: onsets

The detector converts the track into a curve of how much energy arrives moment to moment. A kick drum is a sharp spike on that curve. A held pad is flat. Every spike is a candidate onset — a moment something started.

Onsets are not beats. A hi-hat pattern produces four times more onsets than there are beats, and a vocal breath produces one that is not musical at all.

Step two: tempo

So the detector looks for the *interval* that explains the most onsets. If spikes keep falling 0.5 seconds apart, the tempo is 120 beats per minute and the beats are every 0.5 seconds. This is why detection is confident on a four-on-the-floor track and hesitant on rubato piano — there is no repeating interval to find.

The two ways it goes wrong

SymptomCauseFix
Cuts feel twice as fast as the musicDetector locked to the hi-hat, not the kickHalve the density — cut on every second beat
Cuts land between the beatsLocked to the off-beat, a half-interval outShift every marker by half a beat, or re-detect
Cuts drift in the second halfThe track changes tempoRe-detect, or move the drifting markers by hand
Almost no cuts foundAmbient track with no transientsPick a more percussive song, or place cuts manually

None of these is a failure of the idea. They are the four places where a human ear still beats a curve, and all four are a few seconds of work on a waveform — which is why any tool worth using lets you edit the detected beats rather than only accepting them.

What good detection gives you

  • Frame accuracy — the timestamps are read from the signal, not tapped by a person with reaction time.
  • Repeatability — the same track gives the same beats today and next month, so a series stays consistent.
  • Density control — with every beat known, cutting on every second one is a choice rather than more work.

Easy Reels runs detection on every track you upload and shows the result on a waveform you can edit. Read how to cut a video to the beat for what to do with those cut points once you have them.

Read next