Acoustic Invariants for Vowels
Robert F. Port, Terrance M. Nearey · The Journal of the Acoustical Society of America · 1974
In seeking acoustic parameters that would be invariant with respect to phonetic quality regardless of sex and vocal tract size, we reexamined the data of Peterson [J. Speech Hearing Res. 4, 10–29 (1961)]. He obtained measurements of four formants for 10 steady-state English vowels and two non-English vowels by several men, women, and children. The subjects in consultation with Peterson imitated as closely as possible a set of his reference vowels. A variety of measures were made of these data in both hertz and technical mel scales including the differences between formants, the averages of approximating formants, and various ratios and cross ratios. When examining a single vowel in isolation across all subjects, reasonably good acoustic invariants were found only for [i] and [a]. For many other vowels, much better invariants were found when each vowel was considered in relation to other vowels of the same speaker, that is, by an analysis termed “self-normalization” by Gerstman [IEEE Trans. Audio Electroacoust. AV-16, 78–80 (1968)].