Computational analysis of expression in violin performance

Unrestricted This thesis presents computational methods for the quantitative description and analysis of expressive strategies in violin performances, and applies these methods to eleven commercially available audio recordings of the Andante movement from Bach's Sonata No. 2 for solo violin. First, we develop a methodology for the extraction of beat-level tempo and dynamic data from audio recordings: tempo is extracted manually, while dynamics are obtained using a psychoacoustic loudness model and a perceptually optimized smoothing algorithm.; Next, we analyze tempo and dynamic means and ranges over musical segments of differing scales to identify trends in expressive strategies. We then propose the Local Maximum Phrase Detection method, which quantifies a performer's interpretation of phrase structure by associating each interpreted phrase with a local maximum in the dynamic curve. Throughout, we compare individual expressive and grouping strategies, and examine the similarities and differences between tempo and dynamic variations.