Experimental methods for speech synthesis by rule

Abstract

An experimental system for synthesis by rule has been developed, consisting of a parallel-resonance synthesizer, a laboratory computer, and two computer programs. To drive the synthesizer, values for each of its parameters must be specified at appropriate intervals. The "Rules" program (an elaboration of earlier work done with Holmes and Shearme at the Joint Speech Research Unit) accepts an input string representing an utterance phonemically, and an auxiliary input consisting of rules in tabular form for synthesis of the segmental and prosodic phonemes of a particular dialect. From these inputs, the sets of parameter values for the utterance are calculated. The "Executive" program reads the parameter value sets into memory and transmits them to the synthesizer. This program also enables the variation of the transmission rate editing of the stored values, disk storage and retrieval of the parameter value sets for a large number of utterances and synthesis of a sequence of utterances.