Granular synthesis was first suggested as a computer music technique for producing complex sounds by Iannis Xenakis (1971) and Curtis Roads (1978) and is based on the production of a high density of small acoustic events called 'grains' that are less than 50 ms in duration and typically in the range of 10-30 ms. Typical grain densities range from several hundred to several thousand grains per second where the grain itself may come from a wavetable (e.g. sine wave), FM synthesis or sampled sound. Such high densities of events made the technique difficult to work with because of the large amount of calculation required, and therefore until recently few composers have experimented with it. Using a digital signal processor controlled by a microcomputer, Barry Truax implemented the technique with real-time synthesis in 1986 and incorporated it within an interactive compositional environment, the PODX system, at Simon Fraser University. This technique was exclusively used to realize his work Riverrun.
High and low frequency grains
What is most remarkable about the technique is the relation between the triviality of the grain (heard alone it is the merest click or 'point' of sound) and the richness of the layered granular texture that results from their superimposition. The grain is an example of British physicist Dennis Gabor's idea (proposed in 1947) of the quantum of sound, an indivisible unit of information from the psychoacoustic point of view, on the basis of which all macro-level phenomena are based. In another analogy to quantum physics, time is reversible at the quantum level in that the quantum grain of sound is reversible with no change in perceptual quality. That is, if a granular synthesis texture is played backwards it will sound the same, just as if the direction of the individual grain is reversed (even if it is derived from natural sound), it sounds the same. This time invariance also permits a time shifting of sampled environmental sound, allowing it to be slowed down with no change in pitch. This technique is usually called granulation.
Key to all granular techniques is the grain envelope. For sampled sound, a short linear attack and decay prevent clicks being added to the sound. Changing the slope of the grain envelope, in classic microsound practice, changes the resulting spectrum, sharper attacks producing broader bandwidths, just as with very short grain durations.
Common types of granular synthesis are:Quasi-synchronous granular synthesis:
The top diagram shows a grain stream of equal duration grains, producing Amplitude Modulation with grain durations less than 50 ms. The bottom diagram shows 3 grain streams with variable delay time between grains, the sum of which resembles asynchronous granular synthesis.
Asynchronous granular synthesis:
Grains are distributed stochastically with no quasi regularity.Pitch-synchronous granular synthesis:
Overlapping grain envelopes designed to be synchronous with the frequency of the grain waveform, thereby producing fewer audio artifacts.
For further information of frequency-time methods, see this conference paper. Sound examples are provided on the author's DVD-ROM. You can also watch a video lecture on this subject here.
Roads, C. (2001) Microsound. Cambridge, MA: The MIT Press
De Poli, G., Piccialli, A. & Roads, C. (1991) Editors. Representations of Musical Signals. Cambridge, MA: The MIT Press
Gabor, D. (1947) Acoustical quanta and the theory of hearing. Nature, 159(4044), 591-594
Truax, B. (1988) Real-time granular synthesis with a digital signal processor. Computer Music Journal, 12(2), 14-26