The long-term objective of the research plan is an easy-to-use system for writing rules for speech synthesis, and, in addition, a system for generating high-quality synthesized English from a phonetic and prosodic representation for different speakers and speaking styles. The proposed system has application in research and teaching on normal and disordered speech and in aids for the visually impaired. The system uses a new approach in which the rules specify the time course of a compact set of higher-level parameters related to articulation. These parameters control a flexible Klatt synthesizer with settings that can be adjusted for different speaker characteristics. The rules operate on a linguistic description of an utterance (an annotated sequence specified in terms of a hierarchical arrangement of segments, features, syllables, words, phrases) and incorporate speech knowledge through a declarative approach based on an object-oriented methodoloy. The rule system consists of descriptions of an utterance in terms of an ordered sequence of landmarks which are of three types (vocalic, glide, and consonantal). The higher-level parameter tracks are derived from these landmark sequences. Two products are proposed: (1) a synthesis-by-rule system for English, and (2) an experimental rule-development system for speech synthesis.
Two products are expected to emerge from this project. One is a speech synthesis-by-rule system that is the major component of a text-to-speech system, such as reading machine for the visually-impaired. Such systems are also expected to increase in commercial importance in providing speech response to data base interrogation. The other product is a system that can be used by researchers to develop rules for speech synthesis for other languages, speech styles, and other specialized applications.
