Chapter Four: Synthesis

12. Granular Synthesis and Granulation of Sampled Sound

There is often little distinction made between the terms granular synthesis and granulation of sampled sound, but we will treat them somewhat separately, with less attention payed to totally synthetic granulation, as it is used less and less these days.

Granular synthesis was first discussed seriously as an outgrowth of quantum physics, where researchers such as Dennis Gabor examined the concept of reducing sound to its most basic building blocks he called sound quanta in the late 1940's. Gabor interestingly developed a rotating tape recorder head that, in essence, could break down the recorded material into small bits for time and pitch manipulation. Even before computers became capable of doing the large number of calculations necessary to generate the massive amounts of data necessary, Iannis Xenakis both discussed and used these atomic-level sound quanta, often an enveloped simple waveform like a sine wave, or simply a bell-shaped sound impulse called a wavelet to create statistically-shaped "density clouds," as he discussed in his book Formalized Music. By the time computers were ready to lend significant control to the process, leaders in the technique emerged in Curtis Roads and Barry Truax.

Truax was able to use his PDP minicomputer to control the grain production and results in real time. The most notable piece to emerge using granular synthesis is Riverrun (the first word of James Joyce's Finnegan's Wake). At the heart of the technique is the enveloped grain, a short burst of sound that could be 10-100 milliseconds long. These grains could be produced at a very fast rate and density, for example between 100 and 2,000 grains a second. By controlling the grain frequency, grain duration, grain amplitude, envelope shape, more cohesive pitched sounds can be produced when the grains have similar characteristics, and more random noise-like textures can be created when they have higher variability, all controllable via pre-programing or in real time. Listen to Riverrun here (you may need to search if the URL changes).

Of seeming greater interest in recent decades is the granulation of real-world sampled sound. This technique was used with fascinating results in Paul Lansky's Idle Chatter series, circa 2011, along with several other techniques. Many DAW's, plug-ins, apps, and built-in or add-on objects for synthesis languages provide easy access to the technique. The video below provided a starting point for how the technique works and what some basic parameters are. Granulation of sound is essentially an automated form of amplitude modulation, whereby portions of a larger sound file are multiplied by an envelope over and over again to form what are called grains. We will examine the parameters involved below.

Click video image to play/pause

Anatomy of Granulation of Sampled Sound

A sound file (or real-time sound) is read into a buffer. The composer can then select which region of the sound file to granulate, including the entire file if they wish. The process will be limited to the bounds of that selection, though it is not uncommon to change those bounds during the process. And from here, we list the primary control elements found in most granulators.

  • Grain Envelope: The composer can usually select the shape of an envelope to use. These may be a simple curve, a rectangle, but most often one with one or both sides sloped at various angles. At high rates of production, the grain shape will make a difference in the frequency content produced, and in general, a steeper shape will create a brighter sound.
  • Traversal Rate and Direction: A grain pointer moves through the selected region at a specified rate and also direction. The pointer determines the start point of the grain envelope. It is possible to move through more slowly than the real sound, or more quickly, and to move forwards or backwards. It is also possible to have the pointer stand still in a specific spot, in which case, the grain of sound is repeated over and over again. When the pointer reaches an end of the region selection, it will normally jump back to the opposite end of the region. Some granulators allow it to ping-pong back and forth. A common technique is to have a slow traversal rate and a high grain rate which can produce the effect of stretching out sound without transposition much like phase vocoding does.
  • Grain Size, Grain Duration, Grain Length: The length of the grain envelope, usually in seconds or milliseconds.
  • Grain Rate/Grain Onset: The grain rate is the rate at which grains are produced, usually in grains per second. If a grain is produced faster than the preceding grain envelope ends, then they will overlap and sum. The time elapsed between the start of one grain and the start of the next is referred to as the grain onset, the inverse of the grain rate. As shown in the video above, the combination of grain rate and traversal rate forms the hop size, how far into the buffer the next grain is produced. If the traversal rate is 0, then the hop size is 0.

    If the grain production rate is above sub-audio rate (20 grains per second and above), and with no jitter (see below), the following occurs:

    • If the grain duration is less than the time between onsets, we hear the grain rate's fundamental (plus harmonics).
    • If the grain duration is greater than the time between onsets, then the grain waveform predominates.
    • If the grain duration equals the time between onsets, then amplitude modulation occurs with two sidebands, an upper and lower, surrounding each frequency component of the sampled waveform from the sums and differences of the grain rate.
    The style of granular synthesis with a more or less steady grain rate is referred to as quasi-synchronous granular synthesis. In granular synthesis, if the grain rate is set exactly to the frequency of the grain waveform, it is referred to as pitch-synchronous granular synthesis.

Common Granulation Variants

There are certain added bells and whistles that have come to typify most granulation applications. First and foremost is the addition of constrained randomness to the basic parameters, usually as a percentage or rate. The most common name for this randomness is jitter, but they may be called something else relating to randomness. Not all applications provide all possibilities, but the common ones are:

  • Input Jitter: Rather than the grain pointer moving smoothly along, by adding input jitter, it may jostle back and forth horizontally to add randomness to the grain onset points.
  • Output Jitter: Output jitter adds randomness to the regularity with which the grains are produced.
  • Amplitude Jitter: Adds randomness to the amplitude of each grain.
  • Grain Length Jitter: Modifies the grain envelope length with each grain.
  • Stereo location: A really appealing aspect of most granular applications is their ability to place each individual grain randomly in a stereo field. Often the spread can be specified, for example for the grains to be placed randomly only hard left or hard right (called stereo width).
  • Transposition: Will transpose the original pitch.
  • Transposition Jitter: Will cause the transposition to jump to other pitches. A very common characteristic is to randomly jump octaves.
  • Harmonization: Most granulators can produce numerous other simultaneous transpositions with multiple grain streams. Chords, octaves, etc. can be produced and changed during composition.

If you wish to explore granulation of sampled sound on a Mac, John Gibson's Granulator app can be downloaded here. Below is an example of it's use with a variety of the settings discussed above. If you find a spot that interests you, stop the video and examine the settings.