The stochastic (random) models are implemented in practice by creating a probability density function (pdf), which states the probability of an event occurring, as a function of the variable. For instance, one pdf could give the probability of one note occurring as a function of the note value (pitch). In tonal music the notes of the musical scale used would typically have a larger probability. The values of the notes can then be found, in music generation, by using the inverse image formula [15].
As an example, the probability of a note being played has been estimated from a database of folk songs<#_ftn1>[1]. Music generation from simple note probability does not in itself render interesting music, but a conditional probability of the following note, when a note is played, improves the situation. This means that we add the interval probability to the note probability. The probability of an interval is assumed to be independent of the note; therefore the two probabilities can be multiplied in order to create the pdf used to find the next note. In this case, a rather pleasant stream of music is created, but still without enough structure to be really interesting (unless perhaps in relaxation use).
With the help of musicgrams, visualization tools to illustrate the evalution of rhythm, timbre and chroma (Rhythmogram, timbregram and chromagram), chromatic, rhyhtmic and timbral structure are re-introduced by analysis of existing songs, and understanding of human perception.

[1] The Spring 2002 Digital Tradition Folksong Database, <http://www.mudcat.org>http://www.mudcat.org, (1 Nov. 2007)