Spectral analysis of sung vowels. II. The effect of fundamental frequency on vowel spectra

Abstract

One-third-oct bandfilter spectra of nine different vowels, each obtained as the average spectrum of vowels sung by professional male and female singers in nine modes of singing, were obtained at six fundamental frequencies (F0), ranging from 98–880 Hz. For F0=98 Hz both an empirical and a theoretical comparison were made between the representation of vowels in the formant space and the representation in the spectrum space. A first-order correspondence between both approaches was demonstrated. The vowel configurations in the spectrum space showed that, apart from differences in the grand-average spectrum, (1) for F0<220 Hz, vowel configurations were similar for males and females; (2) for F0=392 Hz, the variability in dimensions related to F1 and F2 was smaller than for lower F0 values, and variability related to F1 in male falsetto register was greater than in modal register (vowel spectra were similar in male falsetto register and female singing); (3) for F0=659 Hz, the vowel configuration had shrunk to clusters of front and back vowels; and (4) for F0=880 Hz, vowel differences were only marginal. The relation between the average sound level of the so-called singer’s formant and F0 appeared to be vowel dependent. Up to F0=392 Hz, the singer’s formant was on the average equally prominent for male and female singers, but for higher F0 its level, relative to overall sound-pressure level, dropped for females. Differences between average spectra of sung and spoken vowels are discussed.