James D. Miller
Arnold F. Heidbreder
Central Inst. for the Deaf, 818 S. Euclid Ave., St. Louis, MO 63110
Synthesizers based on vocal-tract models do not allow independent control of perceptually relevant characteristics of spectral envelopes. The goal is to develop a tool that will allow assessment of envelope characteristics in speech perception by both normally hearing and hearing-impaired listeners. Using LabVIEW (National Instruments, Inc.), a software synthesizer is being developed that is designed to allow systematic and independent control of these envelope characteristics: (1) the locations of up to five spectral peaks; (2) the locations of the minimums (valleys) between peaks; (3) the peak-to-valley depths in decibels; (4) the ratio of peak-to-valley bandwidths; and (5) the overall spectral shape imposed on the patterns of peaks and valleys. The resulting envelopes specify the component amplitudes of carriers consisting of sinusoids. The phases of the component sinusoids can also be individually specified. Additionally, so-called ``autofunctions'' adjust ``nominal'' values of the envelope variables in accordance with the pattern of peak locations in a manner designed to be consistent with natural speech. For example, a nominal peak-to-valley depth is increased or decreased based on the logarithmic separation between adjacent peaks. Synthetic vowels and the effects of manipulation of the variables will be demonstrated.