Evaluating Singer consistency and uniqueness in vocal performances

Johanna Devaney

pp. 173-178

in: Tom Collins, David Mérédith, Anja Volk (eds), Mathematics and computation in music, Berlin, Springer, 2015

Abstract

Identifying consistent and unique aspects of performances is an important aspect of modeling performance style. This paper presents a detailed analysis of inter-singer differences and intra-singer similarities using support vector machines to predict singer identity of a performance from pitch, timing, and dynamics performance parameters. The analysis was performed on a dataset of 72 recordings of the first verse of Schubert's "Ave Maria", the dataset consists of 3 a cappella and 3 accompanied performances by 6 professional and 6 non-professional singers.