Toward a computational model of expression in music performance: The GERM model

This article presents a computational model of expression in music performance: the GERM model. The purpose of the GERM model is to (a) describe the principal sources of variability in music performance, (b) emphasize the need to integrate different aspects of performance in a common model, and (c) provide some preliminaries (germ = a basis from which a thing may develop) for a computational model that simulates the different aspects. Drawing on previous research on performance, we propose that performance expression derives from four main sources of variability: (1) Generative Rules, which function to convey the generative structure in a musical manner (e.g., Clarke, 1988; Sundberg, 1988); (2) Emotional Expression, which is governed by the performer's expressive intention (e.g., Juslin, 1997a); (3) Random Variations, which reflect internal timekeeper variance and motor delay variance (e.g., Gilden, 2001; Wing and Kristofferson, 1973); and (4) Movement Principles, which prescribe that certain features of the performance are shaped in accordance with biological motion (e.g., Shove and Repp, 1995). A preliminary version of the GERM model was implemented by means of computer synthesis. Synthesized performances were evaluated by musically trained participants in a listening test. The results from the test support a decomposition of expression in terms of the GERM model. Implications for future research on music performance are discussed.