Generative sound

This is an abstract term which includes many others like algorithmic, procedural and AI sound. These overlap and are often used as ways of talking about mostly the same thing. For a condensed overview see Jarvelainen [17]. In other words all these things are generative, simply because some process generates the sound as opposed to a human composing it. The definition is often given that a generative piece requires no input, or the input is given only as initial conditions prior to execution. Analysing many practical implementations this definition breaks down, but if we stick to it strictly then generative sound is not interactive.

Generative compositions differ from sequenced ones in that a sequenced composition is laid out in advance and does not change. A generative composition happens as the program runs. The philosophically inclined will wonder whether generative compositions are deterministic. In fact if run on a computer they always are, because computers are deterministic. This is not the same question as whether they are predictable. Something can be deterministic but unpredictable. The answer to this very interesting question depends on whether the algorithm is seeded by truly random numbers. Computers cannot generate truly random numbers, but it is fair to say that a generative sequence can be so complex as to be unpredictable by human perception or reason.

It may include non-computational methods, for example Ron Pellegrino's 1972 processes for "Metabiosis" was an installation with lenses sensing air flow changes using light and photoresistors. Strictly it is not algorithmic, not because it doesn't run on a computer, but because the process does not follow the definition of an algorithm [3]. Whether or not you believe installations that harness environmental unpredictability like Pellegrino's are deterministic depends on whether you believe reality is deterministic.