A method of synthesis that involves concatenating a string of many very short samples (termed grains), ranging from a few milliseconds to about 50 ms in length. This method relies on a pschyoaccoustic principle that states that the human ear needs a sound of a minimum duration, about 100 ms, in order for the brain to determine what the sound actually is. Dividing sounds up into shorter samples and then rearranging them can have surprising results. A common example is the dividing of a recording of human speech into grains, and then rearranging the grains in a random order. The result can sound like ordinary speech, but in an unidentifiable language.
To avoid clicks and pops at the boundaries of the samples, an envelope is applied to each one. Because of the short duration of the grains, the envelope acts as a form of amplitude modulation, and itself has an effect on the resulting sound. A related method of synthesis, wave scanning, eliminates the need for the envelopes by having the grain boundaries always meet at the zero-crossing point of the respective signals. However, wavetable scanning requires extensive pre-processing to match up the zero crossing points, and so is usually only done with pre-selected, fixed waveforms.
As of this writing, the only commercially produced synth that uses granular synthesis is the Access Virus TI series. However, many software packages are available for both PC and Mac platforms that will perform granular synthesis.