Second ESCA/IEEE Workshop on Speech Synthesis

September 12-15, 1994
Mohonk Mountain House, New Paltz, NY, USA

From Database To Speech: A Multi-Dialect Relational Database Integrated with the Eloquence Synthesis Technology

Susan R. Hertz (1), Elizabeth C. Zsiga (1), Kenneth J. de Jong (1), Paul Gries, Katherine E. Lockwood

Eloquent Technology, Inc., Ithaca, New York, USA
(1) also Department of Modern Languages and Linguistics, Cornell University, Ithaca, NY, USA

This paper describes a multi-dialect relational database designed to support development of a multi-voice computer program, called Eloquence, for rule-based synthesis of four dialects of American English - General American, Brooklyn, Boston, and a dialect of Alabama. In addition to higher-level linguistic information, the database includes detailed spectral and durational information about formants, voicing, frication, and aspiration for a variety of utterance types in each dialect. From information in the database, multi-tiered utterance representations can be automatically generated for immediate synthesis with the Delta System, the rule development tool being used to implement and test Eloquence. This paper first gives background information about Eloquence and then describes the database, exemplifying how it is being used to derive rules for the language-universal (LU), language-specific/dialect-general (LS), and dialect-specific (DS) components in Eloquence.

Full Paper

Bibliographic reference.  Hertz, Susan R. / Zsiga, Elizabeth C. / Jong, Kenneth J. de / Gries, Paul / Lockwood, Katherine E. (1994): "From database to speech: a multi-dialect relational database integrated with the Eloquence synthesis technology", In SSW2-1994, 45-48.