Third ESCA/COCOSDA Workshop on Speech Synthesis

November 26-29, 1998
Jenolan Caves House, Blue Mountains, NSW, Australia

Designing Prosodic Databases for Automatic Modelling in 6 Languages

C. Grover (1), J. Fackrell (1), H. Vereecken (2), Jean-Pierre Martens (2), Bert Van Coile (1,2)

(1) Lernout & Hauspie Speech Products NV, Sint-Krispijnstraat 7, 8900 Ieper, Belgium
(2) ELIS, University of Ghent, Sint-Pietersnieuwstraat 41, B-9000 Ghent, Belgium 1122/12

We describe the design and creation of prosodic speech databases for 6 languages. The purpose of the databases is to allow derivation of prosody models in order to improve TTS synthesis. The main prosodic variables to model were word prominence, prosodic boundary strength and phone duration. We describe the database structure and contents and the methodology for creating prosodic databases, and we present statistics for the main prosodic variables.


Full Paper

Bibliographic reference.  Grover, C. / Fackrell, J. / Vereecken, H. / Martens, Jean-Pierre / Coile, Bert Van (1998): "Designing Prosodic Databases for Automatic Modelling in 6 Languages", In SSW3-1998, 93-98.