Granular synthesis on hardware: turning three seconds of audio into a whole track

Granular synthesis chops audio into very short fragments, called grains, and replays them in overlapping clouds. Each grain is typically between 1 and 100 milliseconds long with a short fade at each end. Play a few grains per second and you hear stuttering. Play hundreds and the individual grains disappear into a continuous texture whose character depends on the source material and almost nothing else.

Hardware granular boxes and modules have made this cheap and immediate. The trap is that everything sounds impressive for ten minutes and then all of it sounds the same. Avoiding that is a matter of understanding four controls.

Grain size

Short grains — under about 20 ms — lose the identity of the source and start behaving like a tone generator. You hear the grain rate as pitch, and the source becomes a colour rather than a recognisable sound. Long grains, above roughly 80 ms, retain recognisable syllables and transients, giving a stuttered, sampled quality.

The mid range, 30 to 60 ms, is the awkward zone where you hear both, and it is where most default presets sit. Move away from it deliberately in one direction or the other.

Density and overlap

Density controls how many grains are sounding at once. Low density leaves audible gaps and produces rhythmic, pointillist textures. High density produces a smooth wash but also stacks up level fast, because overlapping grains sum. If your granular patch clips or turns into undifferentiated noise, density is usually the culprit rather than the source.

A useful discipline: set density by ear with the source muted in the mix, then bring it back. Granular clouds are much denser in isolation than they seem when playing along with a track.

Position and scan

Position sets where in the buffer grains are drawn from. Freezing position gives a static drone whose entire character comes from that one slice of audio — this is where the famous "infinite pad from one chord" effect comes from. Slowly scanning position plays the source at an arbitrary speed with no pitch change, which is the single most useful thing granular does.

Add a small amount of random spread to position, and the texture stops sounding like a loop. Two to five percent of the buffer is usually enough. More than that and the source turns to soup.

Pitch and the octave trick

Pitching grains up and down is where clouds gain depth. The most reliable move is not a continuous detune but stacking grains at fixed intervals: unison, plus one octave, minus one octave, and occasionally a fifth. This creates a texture with clear harmonic structure instead of a smeared chorus.

If your device offers random pitch, quantise it if you can. Random continuous pitch sounds like a broken tape machine. Random quantised pitch sounds intentional.

Choosing source material

Granular flatters some sources and destroys others.

• Sustained, harmonically simple material — a held pad, a bowed note, a vowel — turns into a beautiful cloud with almost no effort.
• Percussive material works only at long grain sizes, where transients survive.
• Full mixes rarely work. Too many competing pitches, and the result is grey.
• Field recordings are the sleeper choice: room tone, rain, traffic and machinery give textures no synth patch will reach.

Record three seconds of yourself humming a single note into the buffer. It will outperform most sample packs.

Keeping it in a mix

Granular textures occupy a lot of spectrum and have no transient of their own, which makes them sit oddly next to drums. Three fixes that work almost every time: high-pass the cloud at 150 to 250 Hz so it stops fighting the bass; automate its level down during vocal phrases rather than EQ-ing it hollow; and give it a slightly different reverb from everything else so the ear files it as a separate space.

Used that way, three seconds of audio really can carry an arrangement — as the bed underneath the track rather than the thing on top of it.