Comparing Computer Program and Participant Similarity and Emotion Ratings of the Surface Features within Music

Faculty Supervisor:

Dwight Krehbiel

Year of Project Completion:

2009

Student:

Rebecca Buchta

Abstract

A senior thesis project, paper, and presentation People interact with music quite often in their everyday lives, and similarity judgments are one of the key methods of interaction between musical stimuli and human listeners. Research has shown that people use the surface features of music to determine many of their similarity judgments. Computer music analysis programs also often take into account the surface features and sometimes similarity of different musical pieces. The development of computer models of human musical interaction in the form of similarity ratings can help us understand human listening and build better computer programs. Participants in this study listened to fourteen 40-second long songs; two original pieces, six dissimilar pieces, and six similar pieces. Participants rated the pieces based on their emotional responses as well as the similarity of the pieces. Two computer programs, Armonique and PsySound, were then compared for the surface features rhythm, pitch, and timbre. It was expected that all three measures would show similarity ratings in the same direction and that these similarity relationships would also be seen in participant emotion ratings. Correlations between all three sources of similarity judgments were all above .5. Measures of emotion were found to differ significantly between similar and dissimilar songs for both self-report and physiological data. A major goal of this research was to help determine whether Armonique represented a potential computer model for the process of human music similarity judgment. Initial results seem to suggest that it is a possibility. However, more research needs to be done with future versions of Armonique and other programs.