Scale2fit: Disentangling the determinants of vowels' spectral properties using mixed effects models

This paper introduces scale2fit, a framework for evaluating multiple factors affecting vowels’ acoustic properties jointly, which proves useful for vowel normalization and avoiding over-normalization. The acoustic properties of vowels stem from a combination of multiple factors including the intended vowel quality, vocal tract length, coarticulation, and speech rate. However, most current vowel normalization approaches do not distinguish among the different determinants. Scale2fit uses mixed effects models to jointly evaluate multiple factors, avoiding their conflation in vowel normalization. Study 1 shows that the new method excels in a series of comparisons with other normalization methods. Study 2 shows that the proposed framework excels at preserving vowel space reduction that follows from short duration, and is therefore more appropriate for studying vowel space reduction. Finally, Study 3 shows that it is possible to establish variety norms and add speakers incrementally