The influence of spectral prominence on perceived vowel quality

Research indicates that, when the first and second formants of a vowel are separated by less than about 3.5 Bark, perception of its height and some other aspects of its quality is determined by some weighted average of the low-frequency spectrum, rather than by particular harmonic or hypothetical formant frequencies (as is the case with more widely spaced formants). This spectral averaging has been called the center of gravity (COG) effect. Although the existence of the effect is generally accepted, the factors that govern it are poorly understood. One possibility is that the influence of the spectral envelope on perceived vowel quality increases as low-frequency spectral prominences become less well defined. A series of three experiments examined this possibility in: (1) nasal vowels, where the lowest spectral prominence is broader and flatter than that of oral vowels; (2) one- versus two-formant vowels with bandwidths appropriate for oral vowels; and (3) two-formant vowels with very narrow or very wide bandwidths. The results of these experiments show that, when two or more spectral peaks lie within 3.5 Bark of one another, F1 and the centroid (an amplitude-weighted average frequency that estimates the COG in the low-frequency spectrum) roughly determine the boundaries within which the perceptual COG lies; the frequencies of spectral peaks dominate responses when formant bandwidths are narrow, whereas overall spectral shape exerts more influence when spectral prominences are wide. Assuming that all vowels undergo the same processing, it is suggested that vowel quality, particularly height, is determined both by the frequency of the most prominent harmonics in the low-frequency region and by the slopes of the skirts in the vicinity of these harmonics. These two effects are most clearly separable in vowels with poorly defined spectral prominences whose shape cannot be adequately described by specifying the frequencies and degree of prominence of just one or two harmonics, or hypothetical formant peaks.