Piecewise-planar vowel formant distributions across speakers
David J. Broad · The Journal of the Acoustical Society of America · 1981
A previous study [D. J. broad and H. Wakita, J. Acoust. Soc. Am. 62, 1467--1473 (1977)] showing that one female speaker's first three vowel formant frequencies clustered about a two-part piecewise-planar surface is extended to five additional speakers. For each speaker, a similar two-plane representation is found, with the rms spread of the data about the planes ranging between 69 and 103 Hz. The orientations of the planes for the different speakers are similar: The front-vowel planes make an average angle of 10 degrees to the average front-vowel plane, while the back-vowel planes make an average angle of 13 degrees to the average back-vowel plane. Nearly all these departures from the average are significant at the 99% level. The hypothesis of uniform scaling of vowel formant frequencies between speakers must therefore be rejected if it is carried strictly to three dimensions. This is also shown by the positions of the planes. The speakers do, however, group into two almost uniformly scalable subsets. Finally, the third formants of the retroflex vowels for most of the speakers are lower than would be predicted solely from exploitation of low-F3 regions of the piecewise-planar surfaces.