A multidimensional scaling study of esophageal vowels

A multidimensional scaling experiment was conducted to determine the perceptual structure of 11 American English vowels produced by a competent esophageal speaker. Estimates of perceptual distance among these vowels were obtained using a 9-point similarity/dissimilarity scale and were analyzed using an individual differences scaling algorithm (INDSCAL). A three-dimensional perceptual space was produced. The three perceptual dimensions corresponded to tongue advancement, vowel height, and rhotacization. These three dimensions were then correlated with selected bark scale transformed acoustic vowel measurements. The perceptual dimensions 1-3 corresponded most closely to F3-F2, F1-F0, and F3, respectively. Little difference was found between the perceptual structure of esophageal vowels and laryngeal vowels, although it is suggested that the correlation between some of the acoustic measures (such as F0) and the perceptual dimensions may change as a function of individual speaker differences due to postsurgical capabilities (anatomical and physiological) and/or method of voice restoration.