Dissimilarity Measures and Emotional Responses to Music

That people have emotional responses to music is a truism. However, we have little understanding of the ways in which music brings about these emotions. Indeed, we lack decent ways to measure these responses in a quantitative way. As an early step in this area, we devised a listening experiment with a novel response paradigm. Listeners chose from a set of around twenty emotional descriptors, selecting a strength value for each chosen word. Importantly, we did not prevent the listener from selecting conflicting words, or limit her to only one choice. We then used unsupervised machine learning techniques to explore the space of responses. Early results show good agreement with prior studies, but with the potential for more nuanced understanding. We plan to extend this work into considering a broader space of influencing factors on emotional response.