Validity of Automatic Acoustic Workflow

Abstract

By Justine Heider, Speech and Hearing Sciences

Advisor: Carrie Rountrey

Presentation ID: 154

Abstract: Speech intelligibility is a perceptual measure, and although it is currently the gold standard in clinical practice, it is subject to bias and reliability issues. Acoustic measures, by comparison, are quantifiable and correlated with traditional perceptual intelligibility outcomes. Speech intelligibility is also correlated with quality of life and disease progression in Parkinson's Disease (PD). This project aims to compare calculated acoustic values via automatic acoustic workflow to hand calculated values for validity. The acoustic workflow is the result of an interdisciplinary collaboration between computer science students and the CINCI-lab, targeting four acoustic measurements from elicited and spontaneous speech samples taken from healthy volunteers and persons with PD. The analyzed measurements include: F0 interquartile range, F2 slope, vowel duration, and vowel articulation index. Each measurement was chosen based on its significant correlation to elicited speech intelligibility in previous research. From a larger dataset, a random sample was selected for hand analysis via PRAAT and compared to the automatic acoustic workflow. Assuming coding was run correctly and considering human error, we predict hand calculations will be comparable to the automatic measurement, lending validity to the workflow. This is a crucial first step for translation to clinical practice. If validated, the workflow could be used to automatically measure acoustic values in a variety of speech samples, including both elicited and spontaneous speech. A valid workflow that measures acoustic correlates to functional intelligibility may provide quantitative outcome measures related to quality of life and ascertain reliable disease progression benchmarks for PD patients.