Third ESCA/COCOSDA Workshop on Speech Synthesis
November 26-29, 1998
Prosodic modelling is presented as a two-part process. In the first part, predictor variables are identified on the basis of psycholinguistic factors that determine word grouping. In the second part, these and intrinsic-contextual linguistic factors generate predictions for durational and melodic assignment. It is argued that data-driven approaches can be rendered more efficient by introducing and respecting psycholinguistic principles of human prosodic processing. This study summarises thinking developed conjointly in our laboratory since 1993, as well as portions of the second author’s thesis (Zellner, 1998).
Full Paper (with two sound examples linked from within the paper)
Bibliographic reference. Keller, Eric / Zellner, Brigitte (1998): "Motivations for the Prosodic Predictive Chain", In SSW3-1998, 137-142.