Over the summer, Dr. Wilder and I have set up camp in Drexel University’s Expressive and Creative Interactive Technologies (ExCITe) Center in Philadelphia to work on the Isomer Project: a digital humanities research project aimed at teaching computers to listen to music sensitively. Our goals for the summer are to develop both quantitative and qualitative aspects to Isomer’s ability to listen and identify descriptors for music. During my first day at the ExCITe Center, Dr. Wilder showed me examples of production music, or music used in advertising. Upon listening to production music on its own for the first time, I was surprised at how formulaic it sounded. Despite appearing musically clichéd, advertising music provides the perfect means to understand how emotion is transmitted in music, because it is designed to elicit specific emotions in order to create associations with a specific product. Although it was entertaining to listen to the differences between tracks, we then began to analyze exactly how this music is able to achieve its desired effect.
We started our analysis with mood keywords, such as “tension” and “romantic,” associated with each track. Dr. Wilder wrote a program to automatically extrapolate the keywords associated with each track. To better understand the relationships between the keywords, I researched social networking theory in order to navigate the massive amount of data. I found Mark Granovetter’s recently republished article “The Strength of Weak Ties” (1973) extremely useful in understanding how units relate to one another. In the relatively small sample of tracks we used (around 2,000) we found a surprisingly limited number of unique keywords, but an incredibly dense network of connections. I applied several sorting and ranking algorithms intended for social network exploration with the goal of finding the most popular keywords. Some of these included descriptors such as “vocals,” “pop rock,” and “movement.” From there, we chose several prominent keywords, and developed them into “mood-trees.” For example, one of our root words was “electronic.” Tracks that were electronic had several mood branches, such as “electronic-dark” and “electronic-percussive,” each with their own distinct sound and usages. Eventually, we hope to have Isomer listen to and generate its own music.
Dr. Wilder has been an incredible mentor this summer. One of the most interesting and valuable skills I learned while working with him is the way he ties cross-disciplinary elements into our project. With each task, a related field such as business, programming, and music cognition will intersect with our work – and when that happens, he takes the time to explain exactly how. This experience is more like a real-world collaborative business venture than just a research position in a lab, and I am incredibly grateful to have this opportunity.