Workshop on Artificial Intelligence and Music, MUSIC-AI; IJCAI07

""A cooperative approach to style-oriented music composition"
Espí D., Ponce de León P.J., Pérez-Sancho C., Rizo D., Iñesta J.M., Moreno-Seco F., Pertusa A.

These files are selected demos form the performance of the system described in the paper.

This first version of this demo page shows two selected individual generated by the system after being trained by either classical music or jazz sequences. 8 of the 16 bars have been selected for each demo.

For the classical music sample, the sequence has been played using a string patch in a synthesized.

	k	Ng	Ni	Pc	Pm	ws	wn	wm	distrib	n	|V|
classical	40	2500	30	0.9	0.3	0.1	0.2	0.7	BN	3	40

Another sample with classical training set for fitness:

classical2

30

2500

30

0.9

0.3

0.4

0.2

0.4

MN

3

30

For the jazz sample a sythesized marimba has been utilized, but in order for it to be more appealing, a rhythmic section was cut and pasted from a randomly selected latin-jazz midifile. No processing/editing has been performed for this section.

	k	Ng	Ni	Pc	Pm	ws	wn	wm	distrib	n	|V|
jazz	80	4000	40	0.9	0.3	0.4	0.2	0.4	BN	2	100

This sample is shown as output by the system with any addition. It is synthsized with an organ patch:

jazz2

120

4000

40

0.9

0.3

0.3

0.2

0.5

MN

2

140

Legend:

k	number of neighbours utilized for k-NN
Ng	number of generations
Ni	number of individuals
Pc	crossover probability
Pm	mutation probability
ws	weight for shallow statistical fitness
wn	weight for n-words fitness
wm	weight for melodical fitness
distrib.	statistical distribution for the n-words
|V|	size of the vocabulary (number of n-words selected)