Why Does the Vowel Space Area as an Acoustic Metric Fail to Differentiate Dysarthric from Normal Vowel Articulation and what can be Done About It?

Abstract

Background. The vowel space area (VSA) has been used as an acoustic metric of dysarthric vo4 wel articulation, but with varying degrees of success. Here we test the hypothesis that the failure of the VSA to differentiate dysarthric from normal vowel articulation has to do in part with statis4 tical “noise” that is introduced by insensitivity of Euclidean distances that define the VSA to vo4 wel centralisation. Methods. Differences in vowel production between 5 dysarthric young men post traumatic brain injury (TBI) and 5 young men who served as healthy controls (HC) were tested with four acoustic metrics: the triangular VSA, constructed with the first (F1) and second (F2) formants of the vowels /i/, /u/, /a/, and the Euclidean distances (EDs) between the vowels /i/ and /u (EDiu), /i/ and /a/ (EDia), and /a/ and /u/ (EDau) that define the VSA. The formant frequencies of these me4 trics were logarithmically scaled to reduce irrelevant interspeaker variability. Results. The VSA failed to differentiate between the TBI and HC groups, as did the the EDia, and EDau. In contrast, the EDiu effectively differentiated between groups, both statistical4 ly (unpaired t4test, p=0.0174) and in terms effect size (1.88, large). The significant difference was in the expected direction, indicating vowel centralisation and articulatory undershoot in the TBI speakers. Conclusion. The VSA is likely to perform poorly as an acoustic metric of dysarthric speech because of “noise” introduced by Euclidian distances (EDs) metrics that are not sensitive to the articulatory abnormality (in this study, the EDia and EDau). Thus, rather than using the VSA, it