Estimation of speaker height from formant frequencies

Abstract

This investigation aims at establishing a regression function which predicts the height of a speaker from the formant frequencies measured in the speech signal produced by that speaker. Isolated vowels from a male and a female sample population and data on their body height were collected. The formant frequencies of the vowels were measured and regression analysis to speaker height was performed. It turned out that not all formants and vowels are equally suited for this regression. The major requirement for establishing a (significant) regression function seems to be the accurate measurement of formants. The results of the regression analysis are presented for the vowel [ø:], where best results were obtained. Possible explanations are suggested for the failure of earlier studies to find a correlation between speaker height and the acoustic parameters of the speech signal.