Identifying master violinists using note-level audio features

Abstract

The same piece of music can be performed in various styles by different performers. Vibrato plays an important role in violin players' emotional expression, and it is an important factor of playing style while execution shows great diversity. Expressive timing is also an important factor to reflect individual play styles. In our study, we construct a novel dataset, which contains 15 concertos performed by 9 master violinists. Four vibrato features and one timing feature are extracted from the data, and we present a method based on the similarity of feature distribution to identify violinists using each feature alone and fusion of features. The result shows that vibrato features are helpful for the identification, but the timing feature performs better, yielding a precision of 0.751. In addition, although the accuracy obtained from fused features are lower than using timing alone, discrimination for each performer is improved.