Third ESCA/COCOSDA Workshop on Speech Synthesis November 26-29, 1998 Jenolan Caves House, Blue Mountains, NSW, Australia

Estimation of Articulatory Parameter Trajectory from Speech Acoustic Dynamics

C. Silva (1), S. Chennoukh (2)

(1) Departamento de Electrónica Industrial, Universidade do Minho, Portugal
(2) Center for Computer Aids for Industrial Productivity (CAIP), Rutgers University, Piscataway, NJ, USA

This research aims to perform articulatory analysis as a basis for low bit-rate speech coding. The classical approach consists of gathering a large set of acoustic and articulatory vector pairs in a codebook. Then, based on some criteria, the non-uniqueness of the articulatory trajectories is solved using a dynamic optimization procedure. An articulatory codebook requires a model capable of generating shapes for all possible speech sounds. This paper reports a new approach for incorporating the tongue tip on Ishizaka's vocal tract area function model in order to reproduce more details of certain classes of consonants. Results are reported on the search for an optimized articulatory codebook using different methods of model parameter sampling.

Full Paper

Bibliographic reference.  Silva, C. / Chennoukh, S. (1998): "Estimation of Articulatory Parameter Trajectory from Speech Acoustic Dynamics", In SSW3-1998, 237-242.