loudness.ai
Music analysis / Tempo

Why does a BPM finder show half or double the tempo?

One app says 89 BPM, another says 178. Neither has to be broken: most songs can be counted at more than one speed. We tested automatic tempo detection on 1,659 songs with known tempos to see when this happens and how to choose the number you need.

A song can have two honest tempos

Tempo is the number of beats per minute, but a song has beats at several levels at once. The kick and snare might land 89 times a minute while the hi-hat or a strummed guitar moves twice as fast. Tap your foot to the slow groove and you count 89; nod along to the hi-hats and you count 178. Both describe the same song.

Despacito is a familiar example. It is usually listed at 89 BPM, and detectors that follow its faster rhythmic layer report 178. Drum and bass shows the same thing the other way round: producers set it at around 170–175 BPM, while many listeners feel a half-time groove near 87.

So a reading of half or double the expected tempo is usually a choice of counting level, not a failed measurement. The useful question is which level you need: the one your DJ software, your metronome, your collaborators or your music library expect.

What we tested

We test the analysis behind our music tools on public research collections whose tempos were set by people. For this guide we used 1,659 songs: 661 electronic dance tracks from the GiantSteps Tempo collection, whose labels were corrected in a crowdsourced listening study, and 998 songs in ten genres from GTZAN, a collection widely used in music research, with the tempo annotations published for it.

A reading counted as right when it was within 4% of the published tempo, the tolerance commonly used in tempo research. We also recorded when it landed on half, double, a third or three times the published value.

ReadingShare of 1,659 songsWhat it usually means
At the published tempo76.4%The detector counted the same beat as the people who labelled the song.
Double11.5%It followed a faster layer, such as hi-hats or a strummed pattern.
Half6.1%It followed the slower backbeat of a fast song.
A third or three times0.6%Usually swing or a triplet feel.
Something else5.4%No simple relation: a changing tempo, free timing or a misleading section.

Put together, 94.6% of readings were the published tempo or a simple multiple of it, and 77% of the readings that missed were exactly half, double, a third or three times the label. When a BPM finder looks wrong, the right number is usually one switch away.

Slow songs read double, fast songs read half

The chance of a half or double reading depended mostly on how fast the song is. Between 100 and 160 BPM, readings matched the published tempo 87–97% of the time. Below 100 BPM, the detection often counted the faster layer; above 160 BPM, it often settled on the slower one.

Loudness.ai
Each bar is one range of published tempos. “Other” includes thirds, triples and readings with no simple relation to the label.
Open full-size chart
Published tempoSongsRead at that tempoRead doubleRead half
Under 80 BPM14228%48%0%
80–99 BPM24753%40%0%
100–119 BPM28287%8%0%
120–139 BPM50591%0%3%
140–159 BPM18197%0%3%
160 BPM and over30271%0%26%

Detectors and many listeners lean toward a pulse in the middle of this range. A 70 BPM ballad with steady eighth notes invites a count of 140; a 174 BPM drum and bass track invites 87. Neither the song nor the detector is wrong. They disagree about which layer is the beat.

A practical rule follows: if a BPM finder shows more than 160 for a song that feels relaxed, try the half reading. If it shows less than 100 for a song that feels fast and driving, try the double reading.

Steady grooves are the easiest

Across GTZAN’s ten genres, the steady four-on-the-floor kick of disco gave the most reliable readings and classical music the least. The electronic tracks from GiantSteps matched the published tempo 82.6% of the time, against 72.2% for GTZAN’s mix of genres.

GenreAt the published tempoPublished tempo or a simple multipleMost common miss
Disco94%99%Rare: 2% double, 2% half
Reggae86%99%6% double, 6% half
Rock78%97%17% double
Blues74%94%9% double
Hip hop74%100%25% double
Country72%99%18% double
Jazz69%91%13% half
Metal66%93%15% half
Pop61%98%36% double
Classical49%66%34% with no simple relation

Pop and hip hop almost never missed outright: nearly every reading was the published tempo or exactly double it. Their published tempos sit at a relaxed pulse, while hi-hats and syncopated parts suggest a count twice as fast. Metal and jazz leaned the other way, toward half.

Classical recordings were different. Without a drum kit, and with tempos that breathe and change, a third of them had no simple relation to the published tempo. For music like this, a single BPM number is a rough average at best.

How to choose the right number

The quickest check takes a minute: tap along with Tap Tempo, then compare the result in BPM Finder.

  1. Tap along to the snare or the strongest backbeat for about ten seconds. Your tap shows the level you feel.
  2. Compare your tap with the detected value. If yours is about half or double, keep the detection and switch to the matching reading.
  3. Check the usual convention for the genre in the table below, so your number matches what other people and tools expect.
  4. Keep one convention across a playlist or library. Mixing 87 and 174 BPM labels for the same style makes sorting and searching unreliable.
  5. When the feel matters, write it next to the number, for example “174, half-time feel”, so collaborators read it the same way.
StyleUsually writtenAlso counted as
Hip hop and R&B80–100 BPM160–200 BPM, following the hi-hats
ReggaetonAbout 88–100 BPM176–200 BPM
Pop ballads60–80 BPM120–160 BPM
House118–130 BPMRarely counted differently
Techno125–140 BPMRarely counted differently
Trap130–150 BPM65–75 BPM, its half-time feel
DubstepAbout 140 BPMAbout 70 BPM, its half-time feel
Drum and bass160–180 BPM80–90 BPM, its half-time feel

These ranges are common conventions, not rules. For DJ mixing, half and double tempos still line up: an 87 BPM track can be mixed beat for beat with a 174 BPM track, with every second beat of the faster track landing on a beat of the slower one.

When the number is wrong in another way

About one reading in twenty had no simple relation to the published tempo. These misses were most common in classical recordings, then jazz, metal and blues, which points to the usual causes: tempo that changes or breathes, live playing that drifts, swing that can be counted in threes, and long passages without drums.

  • For a song that speeds up or slows down, treat the BPM as an average and tap the section you will mix or edit.
  • For swing, count each long-short pair as one beat. That is the usual convention, and it avoids readings of a third or three times the tempo.
  • If a song opens with a long free introduction, measure from the first full groove.

In every case, measure the part you care about: trim a steady verse or chorus with Audio Cutter and analyse that section on its own.

Reading the result in BPM Finder

BPM Finder shows the detected tempo together with its half and double readings, so you can switch to the level you hear without measuring again. It also shows the key and its Camelot code, so you can plan a harmonic mix in one step.

Want to know how far you can trust the key as well? Read how accurate automatic song key detection is, or check your keys on the Camelot Wheel.

Sorting a library by style too? Read what a genre finder’s confidence score really means.

References

GiantSteps Tempo: Knees et al., “Two data sets for tempo estimation and key detection in electronic dance music annotated from user corrections”, ISMIR 2015, with corrected tempos from Schreiber and Müller, “A crowdsourced experiment for tempo estimation of electronic dance music”, ISMIR 2018.

GTZAN: Tzanetakis and Cook, “Musical genre classification of audio signals”, IEEE Transactions on Speech and Audio Processing, 2002. Tempo annotations: the GTZAN tempo and beat annotations.

Scoring: Gouyon et al., “An experimental comparison of audio tempo induction algorithms”, IEEE Transactions on Audio, Speech, and Language Processing, 2006.

Known limits of GTZAN: Sturm, “The GTZAN dataset: Its contents, its faults, their effects on evaluation, and its future use”, 2013.

Sign in to Loudness.ai