oa A Clustering Analysis of Renaissance Polyphony Using State-Space Models

In this paper, we propose a novel clustering method to represent and compare works of symbolic music, in particular the music in the database of the Josquin Research Project. We also study the results of our methods on the database as a whole, as well as the results for some specific cases. The methods that we propose revolve around modelling the activity of the twelve pitch classes (i.e., chromae) over time. We suggest that by modelling this activity, we can capture the harmonic progressions in the music. By converting the models to a vector representation, we can analyse the works in the dataset using clustering techniques from machine learning. These techniques can be used to group, visualize, and classify works in the corpus. This way, we suggest attributions for insecurely attributed pieces in the dataset, based solely on the harmonic progressions contained in the songs. We also apply our final classification models to the unica of the Leuven Chansonnier, to give the first suggestions for the attribution of these works from the world of computational musicology. We then compare our methods to some existing methods for analyzing symbolic music, and find that our method is slightly worse at identifying composers of Renaissance works. This result was not completely unexpected, as the features we extract are based only the musical progressions present in the music. Finally, we use the best vector representation we devised together with unsupervised dimensionality reduction algorithms to create a scatterplot visualization of the entire Josquin Research Project dataset.