Three applications of analysis-by-synthesis in music science

Abstract

The article describes how my research has applied the analysis-by-synthesis strategy to (1) the composition of melodies in the style of nursery tunes, (2) music performance and (3) vocal singing. The descriptions are formulated as generative grammars, which consist of a set of ordered, context-dependent rules capable of producing sound examples. These examples readily reveal observable weaknesses in the descriptions, the origins of which can be traced in the rule system and eliminated. The grammar describing the compositional style of the nursery tunes demonstrates the paramount relevance of a hierarchical structure. Principles underlying the transformation from a music score file to a synthesized performance are derived from recommendations by a violinist and music performance coach, and can thus be regarded as a description of his professional skills as musician and pedagogue. Also in this case the grammar demonstrates the relevance of a hierarchical structure in terms of grouping, and reflects the role of expectation in music listening. The rule system describing singing voice synthesis specifies acoustic characteristics of performance details. The descriptions are complemented by sound examples illustrating the effects of identified compositional and performance rules in the genres analysed.