The Perception of Vowel Formants

Abstract

Synthetic vowel sounds, formed by electrical excitation of resonant circuits by a regular series of impulses, have been used as stimuli in a group of psychophysical experiments. The experiments include measurements of just noticeable differences (JNDs) in formant frequency and in band width for the synthetic sounds with one and two resonances, and with resonance band widths comparable to those found in measured vowel spectra. Average JNDs in formant frequency and band width at 1000 cps are 17 cps and 25 cps, respectively. The results are used to establish a frequency scale for the representation of vowel formants. Equal distances on this scale correspond to an equal number of JNDs in formant frequency. The proposed scale is approximated more closely by a linear frequency scale than by a mel scale. The approximate number of JNDs between pairs of vowels (arranged on a scale from front to back vowels) is evaluated, and is in the range 15–20 on the average. In general, the data indicate that frequency analysis of vowels using a filter band width as narrow as 45 cps provides more information than the perceptual processes can resolve.