Computer Aided Composition System with Interactive Selective Population Climbing

In this work, we are developing the computer aided composition system. This system aids a person, which knows cellphone or background music of home page or software. This system is implemented with the interactive selective population climbing. We suppose that it is appropriate to fetch the favorite melody by hill climbing, because we believe that the function about favor in the human can express the proximate single-peak function for acceleration. However, the user´s favorite melodies are sometimes changed, as the user´s favor becomes clear. Thus, the system selects a real parent from many parents in proportion of these evaluation values in order to fetch the favorite melody during keeping variety interatively, whenever the system generates neibour solutions of the real parent. If the value of neibour solution is higher than one of the real parent, the system exchanges the neibour solution from the real parent. The composing model generates the neibour solutions with randomwalk or most likelihood method in order to change the image of the real parent a little. In the experiment, the melody of 8 measures were composed by evaluating about 30 melodies and improving about 20 melodies. We assume that my system can compose the melody by less burden than other systems.