Learning Style-Specific Rhythmic Structures

The goal of computational music modeling is to construct models that capture the structure of music. We present our work on learning style-specific models for the rhythmic structure on a single line of music. The task by which the model is evaluated is to predict the duration of the next note given the sequence of durations leading to that note. We construct several different models that we train using works by a given composer (Palestrina), and assess the success of our models by looking at the prediction accuracy on unseen works by the same composer. We show that introducing style-specific musical knowledge improves the predictive ability of our models.