BPM Finder: Tap Tempo or Detect BPM From a Song

About this tool

This BPM finder gives you the tempo of a song two ways. Tap along with the beat and it averages your taps into beats per minute, with a measure of how steady you were. Or choose an audio file and the page detects the tempo itself, something web tools rarely offer and that usually needs a desktop app. The file is decoded in your browser with the Web Audio API and never leaves your device. A third panel turns any BPM into delay times in milliseconds for setting echo, reverb pre-delay or LFO rates in sync with the music.

How to use the BPM finder

1. Tap tempo: play the song and tap the big button, the space bar or the T key on each beat. The reading settles after eight or more taps. Stop for 2 seconds and your next tap starts a fresh count.
2. From a file: click the file box and pick an MP3, WAV or other audio file. The page decodes it, analyzes up to the first 10 minutes and shows the tempo with half-time and double-time readings.
3. Check the result: if the detected number feels twice or half as fast as the beat you nod along to, use Halve tempo or Double tempo, or tap along to confirm.
4. Delay times: the latest tempo fills the delay table automatically. Edit the BPM box to use any other tempo, then copy the table.

How tap tempo works

Each tap records a high-resolution timestamp from performance.now(). The tool keeps the gaps between your last 17 taps (16 intervals), so the reading is a rolling average that follows tempo changes. The tempo is BPM = 60,000 ÷ average interval in ms. Stability is the sample standard deviation of those intervals, shown in milliseconds and as a percentage of the average interval. A small number means your taps were even; a large one means the average is less reliable. A gap longer than 2 seconds resets the count, so you can start again without pressing anything.

How BPM detection from a file works

The page reads the file with File.arrayBuffer() and decodes it with decodeAudioData() in an offline audio context at 22,050 Hz; MDN notes that the decoded audio is resampled to the context's sample rate. No network request is made with your audio. The tempo estimate then follows the classic onset-strength and autocorrelation approach described in Dan Ellis's paper Beat Tracking by Dynamic Programming (2007), in simplified form:

1. Onset energy. The channels are mixed to mono and cut into frames of about 10 ms with a 5 ms hop. For each frame the tool takes the log-compressed energy and keeps only increases from the previous frame, since a new drum hit or note makes the energy jump.
2. Clean-up. A one-second moving average is subtracted so that steady loud passages do not dominate, and the curve is lightly smoothed.
3. Autocorrelation. The curve is compared with delayed copies of itself for delays up to 8 seconds. Delays that line up many hits score highly.
4. Tempo score. Every tempo from 60 to 200 BPM in 0.1 BPM steps is scored by averaging the autocorrelation at each multiple of its beat period, then refined to 0.005 BPM.
5. Octave choice. A pulse at 140 BPM also lines up on every other beat, which reads as 70 BPM, so scores are weighted toward 120 BPM with a bell curve on a log scale, 0.9 octave wide, the same center and width as the refined tempo weighting in the Ellis paper. If the off-beats turn out to be about as strong as the beats, the doubled tempo is chosen instead.

The beat strength figure is the average normalized autocorrelation at the chosen beat period: near 1 for a metronome, lower for music with a loose or ambiguous pulse.

Worked example

The demo button builds a 20-second click track in memory at exactly 120 BPM and runs the same analysis, which is what you see in panel 2 when the page loads. The detector reports 120 BPM, with a half-time reading of 60 BPM and a double-time reading of 240 BPM, which is outside the 60 to 200 BPM search range. The delay table then shows a quarter note of 60,000 ÷ 120 = 500.0 ms, an eighth note of 250.0 ms, a dotted eighth of 500.0 × 0.75 = 375.0 ms, a dotted quarter of 750.0 ms and an eighth-note triplet of 500.0 ÷ 3 = 166.67 ms.

For tap tempo, suppose your last eight gaps were 512, 498, 505, 490, 503, 497, 510, 495 ms. The average gap is 501.25 ms, so the tempo is 60,000 ÷ 501.25 = 119.7 BPM. The standard deviation is 7.6 ms, about 1.5% of a beat, which is steady tapping.

Common mistakes

• Half or double time. A 160 BPM track can feel like 80 BPM if you tap on the snare. Neither is wrong; pick the one your project or DJ software expects.
• Tapping too few beats. With two or three taps, one late tap moves the result a lot. Tap for at least eight beats and watch the stability figure.
• Trusting detection on complex music. Rubato, tempo changes, swing, live drumming and beatless intros can all mislead an autocorrelation detector. It reports one average tempo for the file, not a tempo map.
• Mixing up note values. Delay plugins often label a dotted eighth as 1/8D and a triplet as 1/8T. Check which one your plugin expects before typing in milliseconds.

Use cases

• DJs and producers checking a track's tempo before beatmatching or warping it.
• Setting a delay pedal, reverb pre-delay or tremolo rate in time with a song.
• Musicians finding the tempo of a recording to practice along with, together with the Timer and Stopwatch for timed practice.
• Runners and dancers picking songs that match a target pace, and video editors cutting to the beat.
• Generating a reference pitch for tuning with the Tone Generator, or checking an audio input with the Mic Test.

Related reading

For a longer explanation, read our guide: How to find the BPM of a song.