Real-time speech synthesis by rule

Publisher Website
Google Scholar
Abstract
A real-time synthesis-by-rule program has been designed to run on a 16-bit microprocessor coupled to a formant-based synthesizer chip. This device is called the Klatt-Talk KT-1 speech synthesis system. English sentences are produced from an input format consisting of a linear string of phonemes, stress, and syntactic symbols. The symbol inventory includes 40 phonemes, one stress symbol (used to indicate the main stress of each content word), and 5 syntactic markers (used to indicate the locations of word boundaries, phrase boundaries, and clause boundaries). Synthesis algorithms are based on research described at previous ASA meetings, and include an algorithm for fundamental frequency contour specification described by S. Maeda, rules for predicting segmental durations described by D. Klatt, phonological recoding rules described by V. Zue, consonant-vowel synthesis rules described by D. Klatt, and a formant synthesizer described by D. Klatt. A demonstration tape will be played. Progress toward the realization of a full text-to-speech system based on the Klatt-Talk synthesizer will be discussed.

This publication has 0 references indexed in Scilit: