AIC Seminar Series

We propose a novel harmony generation method based on decision-theoretic planning. Given the soprano line, we automatically generate the other three voices of the four-part harmony. Our method uses factored Markov Decision Processes (MDPs), a mathematical model of planning under uncertainty. Factored MDPs demonstrate great advantages in an environment where the state space is large. We are the first to model music generation using MDPs.

The talk will provide an introduction to harmony generation, our model, and our algorithm. I will talk about the planning algorithm design, simulation and experiments. Examples and evaluation results will also be included.

Liangrong Yi is a Ph.D. student at Computer Science department, University of Kentucky. She is an International Fellow at SRI’s Artificial Intelligence Center in the summer of 2009. Her research interest is decision-theoretic planning, including algorithms design and relevant applications.