Combining Gestural and Audio Approaches to the Classification of Violin Bow Strokes: 10th International Workshop on Folk Music Analysis

This paper details a brief exploration of methods by which gestural and audio based approaches may be used in the classification of violin performances. These are based upon a multimodal dataset. Onsets are derived from audio signals and used to segment synchronous gestural recordings, allowing for the classification of individual bow strokes utilising data of either type?or both. Classification accuracies for the purposes of participant identification ranged between 71.06% and 91.35% for various data type combinations. Classification accuracies for the identification of bowing technique were typically lower, ranging between 53.33% and 77.35%. The findings of this paper inform a number of recommendations for future work. These are to be considered in the development of a principally similar dataset, for the analysis of traditional fiddle playing styles.