Abstract:

Several investigators [e.g., Chistovitch et al. (1979); Carson et al. (1975); Beddor and Hawkins (1990)] report that listeners match an adjustable single-formant to the spectral center of gravity of a two-formant signal, as long as the two formants are separated by 3.5 Bark or less. This suggests that listeners use some form of spectral averaging in processing vowel sounds. Assmann's 1991 results do not confirm the C-O-G predictions. The preliminary work [Feth et al. (1996)] used the double-staircase procedure to confirm the C-O-G effect for two-formant complex sounds using both periodic pulse train and white noise excitation of the filter used to generate the signals. In the work to be reported here, the relationship between the C-O-G effect and an independent measure of spectral resolution is explored. In 1977, Feth and O'Malley used two-component complex tones described by Voelcker (1966) to estimate spectral resolving power. Resolution bandwidths paralleled the growth of the critical bandwidth with center frequency, but were 2.5--3 times wider. For the same listeners, the limit of spectral averaging for the C-O-G effect and the resolution bandwidth using Voelcker tone pairs have been measured. These results will be compared with predictions of an IWAIF model. [Work supported by a grant from AFOSR.]