| Summary: | This study proposes a fine-tuned approach to modeling musical form in Classical repertoires by analyzing pitch-class distributions in symbolic data using machine learning algorithms. Results suggest that sliding-window histograms, which take the temporal component into account, palpably improve algorithms’ performance in assigning form labels to pieces. Pitch-class histograms were extracted from major-mode piano-sonata movements by W. A. Mozart and L. v. Beethoven according to two methods: whole-piece histograms and sliding-window histograms. In the latter method, richer features were obtained by calculating histograms for 9–90 partially overlapping windows per piece. Supervised learning methods, such as Support Vector Machines (SVMs) and Artificial Neural Networks (ANNs), yield a good separation of data points according to three distinct form labels: sonata form, ABA (ternary) form, and variation form. For both models, using sliding-window histograms significantly improves performance with regard to the whole-piece histogram data. Unsupervised clustering into three components employing a Gaussian Mixture Model (GMM), on the other hand, yields no successful results. Finally, we offer an in-depth exploration of our findings, and propose some directions for further research based on sliding-window data.
|
|---|