Drum Pattern Sequencer

A step sequencer asks one question, over and over: at this moment, does this sound play? Sixteen answers make a bar. No blunter description of rhythm has been built into an instrument, and it has outlasted cleverer ones.

The surprise is how much music survives the reduction. A drummer makes thousands of small decisions in a minute about loudness, timing and tone. The grid throws nearly all of them away and keeps a row of switches. What comes out still makes people move, provided a few of the discarded things are put back by hand. This essay is about which few.

A line of switches

Machines that repeat a rhythm are older than the grid. Between 1930 and 1932 Leon Theremin built the Rhythmicon at the request of the composer Henry Cowell, who wanted to hear combinations of rhythms that were too hard to play; it could sound sixteen of them at once. In 1959 a manufacturer released an accompaniment box that made its sounds mechanically from a rotating disc, with a slider for tempo. Both played what their builders had decided in advance.

The idea of stepping through a row of settings came from the first modular synthesizers. Donald Buchla began building his in 1963 at the San Francisco Tape Music Center, and the system that followed included some of the first analogue sequencer modules: two with eight stages and one with sixteen. Each stage held voltages, and a pulse moved the machine from one stage to the next. The composer Morton Subotnick, who worked with it, recalled programming long and complex rhythms by running five stages against thirteen.

Drum machines took a decade to catch up. In 1972 one appeared with a six-row matrix of push buttons on which the player could enter a pattern by hand, and by the end of that decade a microprocessor was storing patterns the user had written. The layout that stuck was the plainest one: a bar cut into sixteen steps, one row for each sound, a button for every crossing.

The skeleton

Most popular music sits in four beats to the bar, and its drumming hangs on a frame that fits the grid almost too well. The kick drum marks beats one and three. The snare answers on two and four, the backbeat. The hi-hat ticks along in even eighth notes above them. In sixteen steps that is a kick on steps 1 and 9, a snare on 5 and 13, and a hat on every odd step.

Small moves on that frame give whole styles. Put the kick on every beat, steps 1, 5, 9 and 13, with a closed hat between the beats and a clap on 5 and 13, and the result is the pattern known as four on the floor. Leave the snare only on beat three and the same tempo feels half as fast. The frame is so sturdy that a clap, a snap or a stamp can stand in for the snare without anyone losing the place.

Played exactly as written, though, the skeleton sounds like a diagram. Every hat is as loud as every other and each one arrives on a ruler line. Three things soften it, and a good sequencer offers all three: uneven timing, uneven loudness, and cycles that do not match.

Late on purpose

Swing is the oldest of these. In jazz, two eighth notes written as equals are played long and short, close to a triplet in which the first note lasts twice as long as the second. Drum machines borrowed the idea for sixteenth notes, and the designer who introduced it to them has described his method plainly: delay the second sixteenth of every pair, and name the amount by the share of the pair that the first note takes. Fifty percent is straight. Sixty-six is the triplet.

The numbers are small. At 120 beats a minute a sixteenth note lasts 125 milliseconds, so a pair lasts 250. At 60 percent swing the second note of each pair falls at 150 milliseconds instead of 125. That is 25 ms late, a fortieth of a second, and it changes the character of the whole bar. The tool on this page counts time in ticks, 960 ticks to a quarter note, so the same delay is 48 ticks. Set the swing to 60, play a row of closed hats, and the lilt is plain to hear.

The interesting settings are the ones between. That same designer says a setting of 54 percent loosens a straight pattern without it sounding swung, and that at 90 beats a minute a groove feels looser at 62 percent than at the full triplet. This is a matter of taste, tempo and style, and nothing in the arithmetic chooses for you.

Research on jazz recordings points the same way. A study of 456 recorded solos found that soloists tend to play their downbeats slightly behind the rhythm section, typically by about 30 milliseconds at medium tempos, roughly 9 percent of a quarter note, while keeping their offbeats together with the band. When the researchers imposed that pattern on piano recordings, expert listeners rated the result as swinging more. They also report that random timing deviations do not have that effect. Feel, on this evidence, is a small and systematic lateness, and scatter does not produce it.

That is worth remembering in front of a humanize control. A little random variation in timing and loudness stops a machine from sounding stamped out, and the tool offers it, driven by a seed so that a pattern plays the same way twice. It will not create a groove that the pattern lacks. The drum machine designer quoted above reached the same view from the other side: get the dynamics and the swing right, he says, and the notes work best exactly in their slots.

Loud, quiet and uneven

Loudness is the second thing the grid forgets. Drummers fill the gaps around a backbeat with ghost notes, very soft strokes, most often on the snare, that push the rhythm along without being heard as beats in their own right. A ghost note is quieter than a main hit and also has a different tone. An accent is the opposite: one stroke leaning out of the line. Three levels per step, ghost, normal and accent, are a crude scale, but they are enough to turn a row of identical hats into a phrase.

The third softener is the oddest. Tell a machine to place a number of hits as evenly as possible across a number of steps and it will often produce a rhythm that people have been playing for centuries. Godfried Toussaint showed that a procedure written by E. Bjorklund for the timing system of a neutron accelerator does this, and that it has the same structure as Euclid’s algorithm for the greatest common divisor. He called the results Euclidean rhythms and wrote them as E(k,n).

Three hits in eight steps, E(3,8), gives x . . x . . x ., which Toussaint identifies as the Cuban tresillo and notes is also widespread in West African music. Five in eight gives the cinquillo. Seven in sixteen gives a pattern he lists as a samba rhythm necklace from Brazil, a necklace being the pattern regardless of where in the cycle you start. His paper lists many more, from Bulgarian folk dances to bell patterns of Ghana, with the sources for each. These rhythms belong to those traditions. The algorithm only describes how evenly they are spaced.

Uneven cycles are the whole idea of polymeter, which is two or more meters sounding at once. On a step grid it costs almost nothing: give one track a shorter loop than the others. A cowbell that repeats every five steps and a rim click that repeats every three will slide against a sixteen-step kick, and the three meet again only after the least common multiple of their lengths. For 16, 5 and 3 that is 240 steps, fifteen bars of four before the pattern truly repeats.

Sounds from nothing, files for elsewhere

None of this needs a recording of a drum. Analogue drum voices were built from very little, and the recipes still work. A kick is a low tone whose pitch drops quickly as it fades, with a short click at the front. A snare is two fading tones for the drumhead plus a burst of filtered noise for the wires. Cymbals and hi-hats come from six square waves tuned so that none is a harmonic of another, filtered until only a bright cluster is left. A cowbell is two such tones through a band-pass filter, struck hard and left to ring.

The tool’s voices follow those outlines with its own numbers, and nothing here will be mistaken for a wooden shell and a brass cymbal in a room.

Timing in a browser needs one piece of care. A page’s timers can run late by tens of milliseconds, so the sequencer wakes every 25 milliseconds and hands each hit due in the next 100 to the audio clock, which starts it at an exact time regardless of what the page is doing.

A finished pattern can leave in two forms. A WAV file holds the sound itself. The format came from Microsoft and IBM in 1991, and its plainest kind is a short header followed by the samples, here two channels of 16-bit numbers. A Standard MIDI File holds no sound at all, only instructions: which note, how hard, and when, counted in ticks of a quarter note. General MIDI fixes which note number means which drum on channel 10, so key 36 is a bass drum and key 38 a snare on any instrument that follows it.

That difference is the last limit worth stating. The MIDI file carries the swing, the nudges and the retriggers as tick times, but the drums it triggers elsewhere will be someone else’s. And the grid, however fine, is still a grid. It holds a pattern, a few levels of loudness and a handful of deliberate delays. The rest of what a drummer does is not in the file.