BPM Finder

It measures the tempo of an audio file on your device and shows you the evidence: every beat it found, drawn on the waveform, and how the tempo moves through the track. Nothing is uploaded.

• Tempo to two decimals. Derived from the beats it actually found, not from a search-grid bin, so the figure is precise enough to beat-match against.
• A beat grid you can export. Audacity label track, CSV or JSON, with bar and beat numbers when there is an accent pattern to number them from.
• A tempo curve. One number is an average. On a live take or a DJ set the curve is what tells you whether that average describes anything.
• An honest refusal. Ambient material and unpitched noise are refused rather than given a number that looks like an answer.
• Live from the microphone. Point your device at a speaker and read the tempo of something you do not have a copy of. It is the same analysis as the file path, on a rolling twelve-second window.

Half-time and double-time are not errors so much as different levels of the same metre. A hip-hop track with a kick every other beat is 85 BPM to one listener and 170 to another, and both are describing what they hear. What is an error is a tool that picks one and hides the other, or worse, folds every answer into a comfortable middle band so a genuine 60 BPM ballad is reported as 120.

• Nothing is folded. The full range from 40 to 250 BPM is reportable, and a slow track stays slow.
• The alternatives carry evidence. Half, double and the two triplet relatives are each scored separately, so the one being offered is a measurement rather than arithmetic on the winner.
• Switching re-spaces the grid. Adopting double-time gives you twice as many beat markers, because that is what double-time means.

A programmed track sits on one tempo and the curve is a flat line. Anything played by people moves — a band pushing into a chorus, a drummer settling after a fill, a record ripped from vinyl — and the single number becomes an average of tempos the track never actually plays. The curve makes that visible instead of leaving it to a confidence percentage, and the steadiness figure beside it is the same information as one number: 100% is a track that never moves, and it drops as the spread widens.

Every beat is written down, not just counted. The grid exports as an Audacity label track — which Audacity imports directly through File, Import, Labels — or as CSV and JSON for a DAW script or a spreadsheet. Measured against click tracks whose beats are at known positions, the markers land within a couple of milliseconds of the transient, which is close enough to cut on.

• Bars and beats when there is a pattern. An accent every three or four beats is found from the audio, and the labels then read 1.1, 1.2, 1.3 rather than 137.
• Nothing invented when there is not. A track with no accent gets numbered beats and no bar claim, because there was no pattern to measure.
• A whole batch as one archive. Analyse a folder and take every grid away in a single download.

A long ambient intro, a spoken introduction before the music, or one track inside a two-hour set will all drag a whole-file measurement towards something that is not the tempo you want. The analysis window sets where to start and how long to measure, and the part of the waveform that was actually measured is shaded, so a result is never quietly about a different section than you think.

The audio is decoded to a single channel, and the spectrum is measured about eighty-six times a second. Where energy rises sharply in any band, something has started — that is an onset, and a track's onsets are where its beats live.

• Onsets, not loudness. Measuring the rise in each frequency band separately finds a hi-hat over a loud pad, which a measurement of overall volume never can.
• A repeating spacing. The onsets are compared against themselves at every plausible beat spacing to find which one repeats.
• Then the beats themselves. A grid is laid over the whole track at once, so a quiet passage is bridged by the passages either side rather than derailing it — and because that grid is allowed to move, a tempo that drifts is followed instead of averaged away.
• The tempo comes last. It is the spacing of the beats that were found, fitted across every one of them, which is why it is precise to two decimals rather than to the nearest search step.

Powered by optimised audio processing algorithms.

• Measure the busiest part. A chorus with drums gives a stronger reading than an intro without them. Use the analysis window to point at it.
• Check the grid, not the number. If the markers line up with the snare, the tempo is right. If they drift across the track, the tempo moves and the curve will show it.
• Tap along when you disagree. The comparison line says whether your tapping is a different reading of the same beat or genuinely something else.
• Low confidence is information. The three parts behind it say which one is weak: soft beats, an ambiguous octave, or uneven timing.

• Audio cutter
• Change audio speed
• Pitch shifter
• Audio metadata editor
• Audio converter
• Loudness meter