Imitating Human Performances to Automatically Generate Expressive Jazz Ballads

One of the main problems with the automatic generation of expressive musical performances is to grasp the way in which human performers use musical knowledge that is not explicitly noted in musical scores. Moreover, this knowledge is tacit, difficult to verbalize, and therefore it must be acquired through a process of observation, imitation, and experimentation. For this reason, AI approaches based on declarative knowledge representations have serious limitations. An alternative approach is that of directly using the implicit knowledge that is in examples from recordings of human performances. In this paper, we describe a case-based reasoning system that generates expressive musical performances imitating examples of expressive human performances.