Granular synthesis is a process where sounds (really signals of any type, but not usually) are chopped up into millions of bits and lumped together to make other sounds.

Of course it's more complex than that. Any signal could be created through granular synthesis. As an extremely complex medium, its control is understandably complex as well. Portions of sounds, named "grains," form the basis for the process. The grains may be manipulated by a variety of variables. These variables are intricately related, and may be manipulated to create sounds of timbre as complex as the operator wishes.

The most fundamental variable is the time it requires for the synthesizer to recognize a change made by its operator. This may be generalized and have an effect on all subsequently manipulated variables, or may be manipulated on an individual basis. With more complex synthesizers, the opportunity arises to map specific functions to rates of change, although this is rarely seen.

The second most basic of variables one may control are the starting point in an original sound file, and the length of the portion of the file used (grains are not the entire portion necessarily, as it may be divided further), and playback rate. Naturally, these variables are not necessarily constant, and may be adjusted as a function of time, as discussed. They may also be randomized within a certain range with either a static distribution or radial distribution of grain attributes. In some synthesizers, the option is present to map a variable to a sine wave, for instance.

The length of the grain and the rate of grain playback may also be manipulated, which allows for the operator to make sounds as complexly layered as desired (a conversation with three people, for instance, may convincingly be made to sound like a room full of people). Conversely, if the period between grain playback is constant and the grain length is set so that their ratio is a simple fraction (e.g., 1/4), a beat may be created. As the source sound file can be from any source, it can be used (and frequently is) to create beats in so-called IDM music, generally of some abstract quality.

The pitch of the grain may be manipulated as well. It may be lowered to a constant value with respect to the original sound, changed with respect to time, mapped to any given function, and randomized. Glissando (the change of a pitch within a set period, in this case the grain) may also be randomized or set to a function with respect to either grain length or time as seen in the source file. The attack (onset) of the sound may be lengthened or shortened with respect to grain length (theoretically source length, as well), as can the decay (end) of the grain. Finally, volume and panning may be manipulated (quite painfully, theoretically, although I haven't really tried) or manipulated.

I cannot understate the potential for granular synthesis to make any sort of vaguely abstract music or sound. I've used it quite extensively myself to make what some people call music. I used granular synthesis to generate these links.

Iannis Xenakis was one of the first people to use granular synthesis. Go type his name into a search engine. Fuck off.