Random effects and the evaluation of the uniform scaling hypothesis for vowels

Abstract

The uniform scaling hypothesis suggests that formant frequencies of vowels of one speaker can accurately predict those of any other speaker of the same dialect by applying a single multiplicative scale factor. Fant (STL QPSR, 2-3, 1-19, 1975) questioned this assumption and presented graphical evidence that scale factors vary between adult men and women for different vowels in ways that are systematically related to their position in the vowel space. However, the issue is complicated by the fact that average vowel datasets may contain multiple sources of variation, including dialect mixture. Statistical evaluation of the uniform scaling hypothesis (or any systematic deviations from it) need to account for multiple sources of variation. In preliminary random-effects analyses of formant data produced by individual speakers from three geographically distinct dialect regions of American English, we found that while relatively modest systematic trends in nonuniformity may exist, their magnitude may be smaller than suggested by earlier work and their assessment may be strongly influenced by other sources of variation within geographical dialect regions. We will present extensions of our preliminary analyses that will include speech data from Texas, Alberta, and Nova Scotia collected in our laboratories.