The Clap

What it is

The clap, and the difference between four hands and one hand four times.

Real hands never land together, so a clap is several hits inside about thirty milliseconds, and a drum machine copies that: four bursts of noise ten milliseconds apart through a bandpass around 1.2 kHz, with a longer burst decaying behind them. Here you can set the spacing from 5 to 40 ms, choose how the bursts are made, and let it play the two and four of a house bar at 124. The keys move the band, a semitone each.

How it works

There are two ways to make the bursts, and they sound alike and measure differently.

A machine like the 909 has one noise source and opens a gate in front of it four times, so each burst is fresh noise. A sampler, or anyone layering a clap from one recording, plays the same burst four times, so each burst is the same noise. The envelopes are identical. The spectra are not, because a sound that repeats itself every T seconds has a spectrum with peaks every 1 over T hertz and zeros between them, and one that never repeats has no such comb.

The page measures both directly. Sliding the sound against itself, the copied clap comes back to itself exactly at 10.00 ms, 0.1% different. The fresh clap never gets closer than 34% at any lag from 2 to 60 ms. Then it reads the spectrum of the burst train at the frequencies where a 100 Hz comb must have its peaks, every multiple of 100 Hz, and where it must have its deepest zeros, halfway between, and averages the difference across the band. The copied clap gives 23 dB. The fresh clap gives 1.

Widen the copies to 20 ms and the comb closes to 50 Hz and deepens to 25 dB. At 40 ms it’s 25 Hz and 26 dB, and the bursts have stopped being a clap and become a flam. The Web Audio graph does it both ways, a looping noise buffer gated four times or one buffer started four times, and rendered offline it repeats at 10.00, 20.00 and 40.00 ms with combs of 20, 26 and 29 dB, and for fresh noise never repeats and has combs of 2, 3 and 2.

What surprised me

A copied clap has a pitch. Not one you’d sing, but a 100 Hz comb 23 dB deep is a stronger periodicity than most instruments have, and it’s there because of how the sample was made, not because of anything in the sound of hands. The fresh version has nothing at that frequency or any other. Every layered clap in every sample pack carries a comb at one over its layer spacing, and it’s audible as a faint tone under the noise, which is the thing people mean when they say a clap sounds “sampled”.

The measurement I got wrong was the first one I reached for. I read the ripple in the spectrum against a smoothed copy of itself and got 14 dB for both claps. One burst of noise, three milliseconds long, has a spectrum that’s 14 dB ragged before anyone has done anything to it, so ripple on its own says nothing about a comb. The comb has a shape, and the shape says where it must be high and where it must be zero. Asking at exactly those frequencies, and averaging over the band, is what made the difference appear: 23 against 1 rather than 14 against 14.

I also tried to count the bursts off the envelope and gave up. A bandpass rings after each hit, and a half millisecond envelope read the ring as a second burst two milliseconds after the first. A longer window fixed that and then read the tail’s own peaks as bursts. The spacing was the setting, not the finding, so the detector came out rather than getting a third patch.

What I would do next

Hands. Record ten people clapping once each, measure the spread of the hits inside each clap, and see whether ten milliseconds is right or whether Roland guessed. Then the same repeat test on the recordings, which should never repeat, and on a commercial clap sample, which I would bet does.

Then the question of how deep a comb has to be before the ear hears it as a tone under the noise, which is a listening test with a number at the end of it.