Brian C. J. Moore
Joseph I. Alcantara
Dept. of Exp. Psychol., Univ. of Cambridge, Downing St., Cambridge CB2 3EB, England
These experiments investigate whether amplitude modulation of ``formants'' allows vowel sounds to be identified. The periodic complex sounds had a fundamental frequency of 100 Hz and a level of 85 dB SPL. Harmonics were added either in cosine or random phase. Harmonics with frequencies below 3000 Hz were equal in mean amplitude. Three pairs of successive harmonics, located at the first, second, and third formant frequency values of six possible vowels, were sinusoidally amplitude modulated. For a 10-Hz modulation rate, vowel identification of cosine-phase stimuli improved with increasing modulation index, m, up to m=0.4, after which it reached an asymptote of about 80%. For the random-phase stimuli, performance was essentially at chance for all modulation depths, despite the fact that the amplitudes of the ``formant'' harmonics were greater than those of ``background'' harmonics at certain points of the modulation cycle. For a 2-Hz modulation rate, performance improved for the random-phase stimuli, but remained below that for the cosine-phase stimuli. Modulating the ``formants'' at different rates or modulator phases made performance worse for random-phase stimuli but had little effect for cosine-phase stimuli. The results are interpreted in terms of the cochlear-filtered waveforms and in terms of perceptual grouping mechanisms.