Synthesis by rule of consonant‐vowel syllables

Abstract

A FORTRAN synthesis‐by‐rule program has been developed for controlling a cascade/parallel formant synthesizer. The rules have been evaluated by generating consonant‐vowel‐/t/syllables involving all phonologically legal combinations of the consonants /PP, BB, TT, DD, KK, GG, CH, JH, FF, VV, TH, DH, SS, ZZ, SH, WW, YY, RR, LL, HH, MM, NN/ with the vowels /IY, IH, EY, EH, AE, AA, AH, AO, OW, UH, UW, ER, AY, OY, AW, YU/. Five phonetically sophisticated listeners transcribed a randomized list of 336 syllables. There was no prior familiarization with the synthesis. Ninety‐five percent of the consonants and 98% of the vowels were correctly transcribed. Synthesis rules are based on a detailed acoustic analysis of the same syllables spoken by three male and three female talkers. The analysis indicates that some consonants require two or three allopones for accurate synthesis (one before the front vowels /YU, IY, IH, EY, EH, AE/, one before the back unrounded vowels /AA, AY, AW, AH/, and one before the rounded vowels /AO, OW, UH, UW, ER, OY/). Frication spectra and plosive burst spectra are largely invariant before each of these vowel groups. Formant transitions for a consonant can be described by a modified locus theory within each vowel group. [Supported in part by the National Institutes of Health.]