Speech Synthesis-By-Rule Using High-Level Parameters

The purpose is to assess the feasibility of developing a speech synthesis-by-rule system based on a new method for controlling formant synthesizers such as KLSYN88. The new method uses a small set (about 10) of high-level (HL) parameters which, in conjunction with calculated aerodynamic quantities, are mapped into values for the much larger set of KLSYN88 parameters. Previous research has shown that the HL method simplifies the specification of speech sounds, yet yields highly intelligible synthesis output. In addition, numerous regularities have been observed for the HL specifications of different types of speech sounds; this may raise the possibility that an efficient set of rules could be defined for synthesizing sound sequences automatically from symbolic input. The proposal seeks support for development of a subset of the rules that would be needed for a complete synthesis-by-rule system in order to evaluate the extent to which there may be increases in the efficiency of the synthesis rules and in the accuracy of the speech representation as a consequence of using HL parameters.