Anatomy of a Synthesized Drum Kit

Every drum you have heard on this blog is made out of two ingredients: tones and noise. Nothing is sampled, nothing is downloaded. The whole kit is a few oscillators, a few filters, and some decisions about how fast things get quiet.

That matters beyond trivia. Once you know a kick is a falling sine and a snare is noise over a tone, you can hear those parts inside real records, and you can fix your own drums by naming what is wrong with them.

Tap any pad. You will hear the drum and see exactly what came out:

An interactive pad grid: tap a drum to hear it and see the waveform it makes. It needs JavaScript. The recipes:

• kick. A sine falling from 160 Hz to 48 in about a tenth of a second, plus twelve milliseconds of highpassed noise for the beater.
• snare. Bandpassed noise around 1.5 to 2.4 kHz for the crack, over a triangle at 185 Hz for the body.
• clap. Three noise bursts about ten milliseconds apart, then a tail, all through one bandpass. Several hands, not quite together.
• closed hat. Noise through a highpass at 7 to 8 kHz, gone in 25 to 55 milliseconds.
• open hat. The same hat with the decay held open, 300 to 650 milliseconds instead of 30.
• low tom. A sine at 110 Hz bending down a fifth over 180 milliseconds, plus a short tick for the stick.
• high tom. The same drum tuned to 190 Hz.
• ride. Six square waves at inharmonic ratios of 130 Hz through a bandpass, ringing about a second. Metal has no neat harmonic series, so neither does this.
• crash. The same inharmonic bank at 180 Hz, with more noise and twice the decay.
• rimshot. Two milliseconds of noise through a very narrow bandpass at 1750 Hz. Almost all filter, almost no source.

The picture under the pads is the real output, rendered from the same code that just played, not an illustration of it.

Tones: kick and toms

A kick is a sine wave whose pitch falls off a cliff. It starts around 160 Hz and lands near 48 Hz in about a tenth of a second, which your ear reads as one thump rather than as a pitch bend, because the drop is far faster than a melody.

Tap the kick and look at the shape: a fast rise, then a long tapering tail. That envelope is most of the sound’s identity. Give the same oscillator a slow rise and it stops being a drum at all.

The toms are the same trick, tuned higher and falling less far. The only thing separating a low tom from a kick is how far the pitch travels and how long you let it ring.

Noise: hats, snare, clap, rim

Noise is every frequency at once. Filter away most of it and what remains has a character.

A closed hat is noise with everything below about 7 kHz removed, gone in thirty milliseconds. An open hat is that same sound with the decay held open. Tap them back to back and read the length in the readout: same recipe, roughly ten times the tail.

A snare is the interesting case, because it is both ingredients at once. The crack is bandpassed noise. Under it sits a short tone at 185 Hz that gives the drum a body and a pitch. Take the tone away and you have a hiss; take the noise away and you have a small sad drum.

The rimshot is the extreme end: two milliseconds of noise through a very narrow filter. Almost all filter, almost no source. What you hear is essentially the filter ringing.

Metal: ride and crash

Cymbals are the hard ones. Tuned instruments have neat harmonic series, where the overtones are whole-number multiples of the fundamental. Metal does not. Its overtones are inharmonic, and that mismatch is exactly what makes a cymbal sound like metal instead of like a bell or a note.

So the ride and crash here are six square waves stacked at deliberately non-whole-number ratios, filtered and left to ring. It is the trick the TR-808 used, for the same reason: filtered noise alone hisses where a cymbal should ping.

Why hitting harder changes more than volume

Switch the pads to Soft and tap the closed hat, then switch to Hard and tap it again. It is louder, but listen past that: it is also brighter and it rings slightly longer.

That is deliberate, and it is how real drums behave. Hit a drum harder and you get more attack noise, more pitch bend, more ring. If velocity only changed volume, a programmed beat would sound like someone turning a knob rather than someone playing.

Every voice here maps intensity to timbre as well as level: the kick gains beater click and pitch, the snare trades body for crack, the hats loosen, the cymbals brighten.

Reading a waveform

The display draws the loudest sample in each slice of time as a bar above and below a centre line, which is what every audio editor shows you. One difference worth knowing: it zooms to fit each sound, in both directions, so a thirty millisecond hat and a two second crash both fill the frame. The length beside the name is the real duration, and loudness you can hear for yourself. Two things are worth reading off the shape:

Attack, the left edge. A vertical wall means a click. A ramp means something softer, like a mallet or a brush.

Decay, the slope down to nothing. Straight down is a gate. A long curve is a drum ringing in a room. Most of what people call the character of a drum machine lives in that curve.