Real Time Algorithmic Musical Variation Using Style

My honors thesis aims to develop a program that could assist with music composition, or even write interesting music on its own. The starting point is a melody and chord progression composed by a musician, which is the same amount of information that a Jazz performer might get with a lead sheet. Then the computer is tasked with writing a harmony to the melody. The harmony is not only based on the melody and chords, but also on what the program user might want to hear. The program can be provided, in real time, with some descriptors of what a user might want to hear next-such as the words bright and happy , and the resultant music will be a little upbeat and in a major key. The program accomplishes this through use of a hidden Markov model, which is a collection of probabilities for selecting which composition rules to use. Composition rules will affect how the harmony is voiced (from block chords to syncopation to arpeggios), the tempo of the theme, the key of the theme, or some other quality of the music. Each word the user inputs changes the probabilities that the next rules will use. Obvious applications for this program are primarily in the entertainment industry. Video games and other electronic entertainment are prime candidates for this system, allowing for unique and fitting music to be generated based on some of the events occurring in the game