Speech Coding and Synthesis Using Parametric Curves

Jesus (1997)

A Thesis Submitted for the Degree of Master of Science by Research to the School of Information Systems
University of East Anglia, Norwich, UK

October 1997

Supervised by Dr. Gavin C. Cawley

Examined by Dr. Andrew Breen and Professor Stephen J. Cox

Accurate modelling of co-articulation, the context-sensitive merging of the boundaries between allophones in continuous speech, is vital for natural sounding speech synthesis. This thesis describes research investigating the use of Bézier Curves to form models of the effects of co-articulation in human speech. The trajectories of line spectral pair (LSP) parameters through each diphone are represented by cubic Bézier curve segments, found using the Levenberg-Marquardt curve fitting method. The Bézier curve is found to provide an effective model of the transitions between a majority of speech sounds, however a more complex model is found to be needed where abrupt or complex transitions are required, for instance during plosives.

Jesus1997.zip (PostScript file, compressed with Winzip 8.0)

Jesus1997.pdf (Acrobat 6.0 file)

Back to Homepage