Two-formant models of vowel perception: Shortcomings and enhancement

Abstract The assessment of two-dimensional models of vowel perception is taken several stages further. Results are, prima facie, not encouraging: the formula proposed by Paliwal, Lindsay and Ainsworth [13] performs poorly at predicting the perceived upper formant (F2') values reported by Bladon and Fant [1]; and the formulae proposed by both these groups of authors are susceptible to an ambiguity error, whereby a calculated F2' can be shown to be associated with more than one vowel quality. However a key to unlock these (and other) difficulties may be found in the notion of auditory spectral integration of vowel formant energy within a broad bandwidth of approximately 3.5 Bark. On this assumption, nonlinearities in perceived vowel quality, of exactly the types observed, would be predicted as vowels fall under, or escape from, an integration skirt. If therefore the broad-band auditory integration hypothesis is built into existing two-formant perceptual models, their prognosis is greatly enhanced.