Bloothooft,G., Hierden, Ellen van, Nijland L. (1994). The perceptual vowel space at high pitches. Voice Conference, Philapdelphia (abstract).


The perceptual vowel space at high pitches


Gerrit Bloothooft, Ellen van Hierden, Lian Nijland

Research Institute for Language and Speech (OTS)
Utrecht University, Trans 10, 3512 JK Utrecht, The Netherlands

ABSTRACT

It is well-known that sung vowels loose much of their intelligibility as the pitch is raised, or, in other words, that there is an increasing mismatch between the perceived vowel identity and the vowel the singer intended to sing. We did an identification experiment with sung vowels, not only to show that correct identification drops dramatically with increasing pitch, but also to study more closely the type of errors that are made by listeners. Stimuli were nine Dutch vowels sung by four female singers at F0 = 220, 392, 660, and 880 Hz, in a loud and soft mode of singing. For each F0 value, the vowels were presented in a random order to ten phonetically trained listeners. This resulted, for each F0 value, in a confusion matrix of intended and perceived vowels. We wanted to present these data in a perceptual space where similar vowel stimuli are close to each other and where dissimilar vowel stimuli are far away. To this end, we hypothesized that intended vowels with the same distribution of responded vowels are perceptually similar, while intended vowels with a very different pattern of responded vowels are perceptually distinct. Using the (dis)similarity of response distributions (by means of the chi-square goodness of fit value) as a distance measure, we were able to construct vowel spaces for each F0 value. At higher pitches, differences between singers proved to dominate differences in intended vowels. Surprisingly, if we considered the most-responded vowel per intended vowel as the perceived vowel, the well-known vowel triangle still emerged, but now with singers associated to vowel types. This implies that listeners have perceptual anchor-points that are related to vowels at high pitches as well, but that these 'vowels' are only very loosely related to intentionally sung vowels, and more strongly relate to the voice timbre type of different singers.