Find the BPM of a track

How the detection works

The file is decoded, summed to mono and reduced to an onset envelope — a measure, eight hundred times a second, of how sharply the energy is rising. Drum hits and note attacks show up as spikes in that envelope; sustained sound does not. The envelope is then correlated with delayed copies of itself, and the delay that produces the strongest agreement is the beat period.

That last step has a well-known trap. On anything periodic, the correlation at twice the beat period is just as strong as at the beat itself, because every beat still lines up. A detector that simply takes the highest peak will therefore report an arbitrary multiple of the real tempo. This one scores each candidate by the correlation at its period and at two, three and four times it, which is the pattern a genuine pulse leaves and a doubled reading does not.

Why half and double time are shown

Because the ambiguity is real, and not only for machines. A drum and bass track at 174 is counted by some people as 87 with a half-time feel, and both are defensible descriptions of the same recording. The same is true of a lot of trap, which sits at 140 or so but is often written as 70.

Where a machine and a listener can disagree, the honest interface shows both. The number chosen as the headline is the one weighted toward the range people actually tap in, around 120 BPM, which is the standard perceptual assumption. If it looks wrong, the alternate is one click away and it is probably what you wanted.

What it is good at, and what it is not

It is reliable on anything with a clear percussive pulse: house, techno, hip-hop, rock, pop, drum and bass. Measured against drum patterns at known tempos, the answer lands within a fifth of a BPM, and on a bare click track within a twentieth.

It is much weaker on material with no transients — ambient, drones, solo legato strings, a cappella singing — because there is nothing for the onset envelope to catch. On that kind of file the confidence reading collapses, measurably: a real drum pattern scores around 0.2 to 0.35 on the internal scale and a sustained drone scores under 0.01. When that happens the page says so instead of showing you a number that means nothing.

It is also unreliable on live recordings that drift in tempo, since it reports a single average rather than a moving one, and on rubato classical performance the idea of one BPM is not really meaningful.

Getting an exact tempo for a DAW

A detected tempo is an estimate of what was played, not a recovery of the project's setting. If a track was produced at exactly 128 and you get 127.9, the true answer is almost certainly the round number — producers rarely choose fractional tempos, so rounding to the nearest whole number is usually right.

Vinyl rips and tape transfers are the exception: they are genuinely off-grid, because the playback speed was not exact. A record mastered at 120 and ripped from a slightly fast turntable might read 120.7, and that is a real measurement rather than an error. If you need it on the grid, correct the speed with the speed tool in pitch-shifted mode, which changes tempo and pitch together the way a turntable does.

Using it to match two tracks

The common job is fitting a sample or an acapella to a project. Find the BPM of both, then change the sample's speed by the ratio of the two — a 90 BPM loop going into a 120 BPM project needs 120 ÷ 90, or 1.333×. In pitch-shifted mode the sample also rises in pitch, which is what a hardware sampler would do and is often musically fine. In pitch-preserved mode the key stays put, which is what you want if the sample is melodic and already in the right key.

If the loop needs to be trimmed to an exact number of bars first, the bar length is shown above, and the cutter takes it from there.

Nothing is uploaded

The whole analysis happens in the page. That matters more here than on a format converter: the files people want tempos for are usually unreleased demos, promos and stems, and there is no reason any of that should be sitting on someone else's disk to have a number counted.