Speech Synthesis Based Upon Fluid Dynamic Principles

The aim of this project is to characterize the fundamental processes by which speech sounds are produced. Advances in speech technology are hampered because of incomplete understanding of these processes. Current knowledge does not successfully bridge the gap between classical approximation of speech as a linear acoustic description, and a full compressible Navier-Stokes (NS) description, which includes not only the sound propagation described in the linear model, but also the non-linear processes which produce the sound. To address this shortfall, the current research integrates three interrelated approaches for obtaining a deeper understanding of speech production: (1) synthesis of speech sounds using a NS solver, (2) experimental validation of the NS code through acoustic and fluid-dynamic measurements on a physical model, and (3) development and implementation of a reduced model of speech sound production. The expected result is speech synthesis of quality and naturalness surpassing that yet achieved, and a new model for speech generation. The results will lead to improved parameterizations for speech synthesis, speech recognition, and low bit-rate transmission.