AI Insight
This study compared three acoustic measures of vowel space to assess speech differences in 28 people with Parkinson's disease versus 28 controls. While all three measures showed group differences, only token-based measures (triangular vowel space area and vowel articulation index) could statistically discriminate between groups and correlate with speech intelligibility, with the vowel articulation index performing best. However, all models showed poor overall predictive performance.
Why it matters
These findings suggest that token-based vowel space metrics, particularly the vowel articulation index, may be useful clinical tools for quantifying speech changes and intelligibility decline in Parkinson's disease patients. This could help clinicians monitor disease progression and evaluate speech therapy interventions more objectively.
Understand the Science
by Defne Abur, Megan Cushman, Courtney J. Dunsmuir, Cara E. Stepp
This study assessed the ability of three acoustic measures of vowel space, two token-based and one trajectory-based, to quantify speech differences in people with Parkinson’s disease compared to controls and correlations with speech intelligibility. Fifty-six speakers (28 people with Parkinson’s and 28 controls) read a custom reading passage containing corner vowels. For token-based measures, the triangular vowel space area (tVSA) in kHz2 and the unitless ratio of vowel articulation index (VAI) were calculated. For the trajectory-based measure, the articulatory-acoustic vowel space (AAVS) in kHz2 was calculated. Speakers also read five unique sentences varying in word length, which were used to collect speech intelligibility ratings. Analyses of Variance revealed that all vowel space measures yielded group differences. Binary logistic regression revealed that only the token-based measures were able to statistically discriminate PwPD from controls, however all models performed poorly (R2 < 0.113). Linear regressions showed that token-based metrics were statistically related to intelligibility and that the VAI yielded the best model for the current sample. The results support that token-based measures are well-suited for quantifying changes in acoustic vowel space and perceived speech intelligibility in people with Parkinson’s. Future work should examine the generalizability of these findings to other motor speech disorders as well as across different speech stimuli.