Speech Synthesis by Rule

Abstract

It has seemed useful for both theoretical and practical reasons to attempt to write rules for synthesizing speech on the basis of the experiments performed by colleagues at the Haskins Laboratories. Because the discrete nature of phonemes as perceptual units is not reflected on the acoustic level, it is not feasible to synthesize speech by putting together prefabricated phonemes. On the other hand, synthesis of syllabic units need not require an inventory of rules comparable with the number of syllables in the language, since the rules can be phrased in terms of the constituent phonemes and combinatorial operations. The rules can be further simplified by writing them in terms of the acoustic features that correspond to the standard phonetic dimensions of place and manner; syllables and words are then synthesized by combining the rules for the classes to which the constituent phonemes belong. An example of speech synthesized by rule will be presented. [This work was supported by the Carnegie Corporation of New York, the Veterans Administration, and the Department of Defense in connection with Contract DA 49-170-sc-2159.]