ABSTRACT
We present results from a decade-long project in the intersection of artificial intelligence, cognitive neuroscience, computer science, and psychology of music. We have extended original research by George Kingsley Zipf to explore connections between power laws (e.g., Zipf’s law) and music aesthetics, the latter in part defined by emotional responses of human listeners. Our results suggest a strong connection between music aesthetics (as perceived by humans) and the complexity or entropy of music (as measured by metrics based on Zipf’s law). We believe this reflects the fact that both music and the human brain are self-similar, and that our measurements quantify shared aspects of this fractal nature. We introduce Zipf’s law and related power laws. We discuss earlier work connecting complexity of artifacts to aesthetics and perceived pleasantness. We provide an algorithmic description of our metrics and identify the various dimensions they measure. We present experimental results, derived with artificial neural networks, which demonstrate the connection between power laws (as captured by our metrics) and music aesthetics (as captured by popularity statistics from a music Web site). We further demonstrate this connection through Armonique, a music similarity engine based on power laws. The aesthetic similarity of Armonique’s recommenda-
tions is assessed through various psychological experiments involving human listeners. These experiments compare Armonique’s music recommendations against human emotional and physiological responses, further demonstrating the connection between power-law metrics and aspects of human emotion and aesthetics.
