Eloquent Technology, Inc. (ETI) proposes to develop speech synthesis software for five dialects of American English, each with male and male and female variations, for use in unlimited vocabulary vocal communication aids. ETI will use rule-based synthesis with an innovative methodology that enables faster and more economical development of high- quality synthetic voices then formerly possible. The rules will be organized into relatively large dialect-universal modules and much smaller dialect-specific ones. To develop modular rules efficiently, ETI will use (1) a novel linguistic model, (2) a multi-dialect relational database, and (3) the Delta System. The linguistic model leads to simpler rules and captures similarities across dialects. The multi- dialect database will contain the linguistic and acoustic information for a large number of utterances in different dialects from which to extract rules. ETI's Delta System provides an interactive environment for exploring rules, and a special programming language for implementing them. Our Delta rule modules will be compiled into a portable format for use in end-products. ETI will document extensively both the rules and the methodology used to derive them, facilitating the efficient and economical development of a wide range of additional voices in the future.
