Just found quite interesting thesis.

Model should be able to encapsulate a sense of composer motivation whilst not relying on a single algorithm for the composition process. As such, this work describes a new framework with the ability to provide a means to generate music from film in a media-driven, distributed, manner. This includes the initial annotation of the media using our new OntoMedia ontology; the mapping of annotated information into parameters suitable for compositional algorithms; the design and implementation of an agent framework suitable for the distribution of multiple composing algorithms; and finally the creation of agents capable of handling the composition of musical elements such as rhythm and melody. In addition, a case study is included which demonstrates the stages of the composition process from media annotation to automatic music generation.

Good to understand that about 90% of all research papers and software is utter crap :-)

You can check it at http://eprints.soton.ac.uk/263924/1.hasCoversheetVersion/thesis.pdf