A Preliminary Computational Model of Immanent Accent Salience in Tonal Music

We describe the first stage of a two-stage semialgorithmic approach to music performance rendering. In the first stage, we estimate the perceptual salience of immanent accents (phrasing, metrical, melodic, harmonic) in the musical score. In the second, we manipulate timing, dynamics and other performance parameters in the vicinity of immanent accents (e. g., getting slower and/or louder near an accent). Phrasing and metrical accents emerge from the hierarchical structure of phrasing and meter; their salience depends on the hierarchical levels that they demarcate, and their salience. Melodic accents follow melodic leaps; they are strongest at contour peaks and (to a lesser extent) valleys; and their salience depends on the leap interval and the distance of the target tone from the local mean pitch. Harmonic accents depend on local dissonance (roughness, non-harmonicity, non-diatonicity) and chord/key changes. The algorithm is under development and is being tested by comparing its predictions with music analyses, recorded performances and listener evaluations.