Automatically measuring biomechanical skills of violin performance: an exploratory study

Abstract

This evaluation study explores how automated movement analysis can be used to catch the biomechanical skills needed for a physically accurate violin performance, maximizing efficiency and minimizing injuries. Starting from a previously recorded multimodal dataset, we compute movement features from motion captured data of five violinists performing three violin exercises: octave shift, string crossing, and a Romantic repertoire piece. Three violin teachers were asked to evaluate audio, video, and both audio and video stimuli of the selected exercises. We correlated their ratings with automatically extracted movement features. Whereas these features are purely visual (i.e., they are computed from motion captured data only), we asked teachers to also evaluate audio because it can be considered as the direct translation of movement skills into another modality. In this way, we can also look at possible relations between evaluation of the audio aspects of the performance and biomechanical skills of violin playing. Results show that the proposed movement features can be partially used to measure the biomechanical skills of the violin players to support learning and mitigate the risk of injuries.