Artificially Intelligent Accompaniment using Hidden Markov Models to Model Musical Structure

For performing musicians, musical accompanists may not always be available during practice, or an available accompanist may not have the technical ability necessary. As a solution to this problem, many musicians practise with pre-recorded accompaniment. Such an accompaniment is fixed and does not interact with the musicians playing: the musician must adapt their performance to match the recording. To synchronise accompaniment with the soloist, it is preferable that an accompanist should be able to follow the musician through the score as they play, rather than the other way around. During performance, musicians may deviate from what is written in the score (either intentionally, by adding their own musical interpretation, or accidentally, by making performance errors). The accompanist should adjust their playing to follow the soloist. This work investigates how an artificial musician can follow a human musician through the performance of a piece (perform score following) using a Hidden Markov Model of the pieces musical structure. The computer musician is designed to interact with the human musician and provide accompaniment as a human accompanist would: musically and in real time. Having successfully implemented this representation, the performances of the resulting artificial accompanists has been evaluated both qualitatively, by human testers and quantitatively, by objective criteria based on that used at the Music Infomation Retrieval and EXchange Conference in 2006. The artificial accompanists can, in general, accompany human performers with a reasonable degree of accuracy. Testing has also raised an interesting reflection on the nature of co-operation between soloist and accompanist, and more generally on the role of the computer musician in ensemble performance