Decomposition of fundamental frequency contours in the general superpositional intonation model

The superpositional approach to intonation models the fundamental frequency (F0) contour as the additive combination of phrase, accent, and segmental perturbation curves. Existing methods for estimating these components depend on specific intonation models, limiting generality. This research develops an algorithm that decomposes arbitrary F0 contours using only the minimal assumptions of the superpositional framework. The algorithm is applied to examine the relationship between F0 shape and perceived prominence and to enable recombinant (multi‑level unit sequence) speech synthesis. By extracting intonational components directly from natural speech, the method supports flexible recombination and analysis. More broadly, this work enables systematic investigation of linguistic influences on intonation and contributes toward richer, more generalizable superpositional intonation models.