Syllable synthesis

Abstract

In an attempt to provide a more direct representation of coarticulatory effects on spectral and temporal patterns of speech than is possible with a segmentally organized algorithm for synthesis by rule, an algorithm now being developed is organized around the syllable. The input is a phonetic description of the utterance in terms of syllable features and a set of rules. The rules determine the translation of the feature description into a description in terms of articulatory influences, such as those of the current vowel, the following vowel, initial and final semivowels, initial and final consonants. Each such influence is represented by a set of target values for formant frequencies and by an exponential function of time; and the targets, the rates of growth and decay of the influence functions, and their timing relative to one another, are specified in the rules. The influences are ordered, and a formant‐frequency value at a given time is determined by iteratively computing the sum of the target value associated with the nth influence, weighted according to its influence function, and the value determined by the first n − 1 influences. This algorithm is a part of a synthesis by rule program written in FORTRAN for the PDP11/45. [Work supported by NIH and VA.]