Learning Adaptive Game Soundtrack Control

In this paper, we demonstrate a novel technique for dynamically generating an emotionally-directed video game soundtrack. We begin with human Conductor observing gameplay and directing associated emotions that would enhance the observed experience. apply supervised learning to data sampled from synchronized input features output emotional direction in order fit mathematical model Conductor's direction. Then, during gameplay, maps state output, which is then music generation module generates emotionally-relevant gameplay. Our empirical study suggests random forests serve well modeling our two experimental genres.