Realism in synthetic speech: Synthesized speech may be intelligible, but it often sounds artificial; researchers are solving that problem

摘要

A table is presented giving representative commercial text-to-speech conversion systems, and one of these, the Prose 2000, is described in detail. Research being carried out to solve the problem of natural pronunciation is surveyed. It is noted that the storage requirements can be reduced by using general rules for pronunciation of morphs, instead of a morph dictionary. The problems that must be overcome in picking the right sound and producing speech electronically are surveyed.