Speech synthesis by hand and by rule for use in perceptual studies

Abstract

A computer-simulated formant synthesizer will be described that provides a flexible method of generating synthetic speech with precisely controlled acoustic characteristics. The output can be used to study systematically the relative perceptual importance of different acoustic cues to phonetic contrasts. Listings of a machine-independent FORTRAN program for specifying synthesizer control parameter values as a function of time are available along with the FORTRAN synthesizer software. Documentation, including preliminary rules for consonant-vowel synthesis, is under development. Since hand synthesis of lengthy material is very tedious, a program useful for the generation of sentence-length utterances to study prosodic variables and segmental perception in sentence contexts is also under development. A synthesis by rule program is used to lay out preliminary specifications for all control parameters. The rule program makes parameter files that can be modified easily by hand before waveform generation. The user can also intervene early in the rule process and, e.g., modify segmental durations before parameter contours are drawn automatically. These programs are valuable additions to experimental techniques such as using pieces of natural speech or vocoder-modified natural speech because it is possible to test perceptual results under conditions where the stimulus is known precisely. [Supported by an NIH Grant.]