Severely speech-impaired individuals need communication devices that provide natural-sounding synthetic voices that are appropriate in terms of sex, age, nationality, and dialect. The creation of such voices has been hampered by high development costs resulting from a lack of flexible software tools and adequate linguistic models for text-to-speech synthesis. Eloquent Technology, Inc. (ETI) proposes a novel modular approach to rule- based speech synthesis intended to reduce development costs for high- quality synthetic voices. The key tool in this approach is ETI's Delta System, a programming language and interactive development environment for developing text-to-speech algorithms. In the modular approach, a single program module (the base module) builds the part of an utterance representation common to all voices in a given language; smaller, independent voice modules produce the variations among the voices. The linguistic model on which our synthesis rules will be based uses Delta's innovative multistream data structure for representing utterances. In Phase I, we will test the feasibility of the modular approach by implementing a base module for English, and voice modules for two American dialects.
