Advancing speech intelligibility assessment: Comparing school-based speech-language pathologists, laypersons, and speech recognition software.

This non-experimental study utilized a convenience sample (N = 86) to investigate the comparability of school-based speech-language pathologists (SLPs), laypersons, and speech recognition software in the assessment of speech intelligibility for children with phonological disorders. Eighty-nine percent of participants were female, and 92% were white. All participants completed a survey comprised of 25 audio recordings of single-word utterances from an AI-generated voice of a four-year-old male with a phonological disorder and were asked to select what word they believed was heard from a pool of 12 options. The survey also included a passage read by the same AI-generated voice, and participants were asked to use a sliding scale to provide a percentage rating of the amount of the passage they understood. Descriptive and summary statistics found that word accuracy scores increased from the computer software, to the laypersons group, to the speech-language pathology group. Group comparisons were conducted to identify differences between groups (SLPs, laypersons, and speech recognition software) for word accuracy scores and percent intelligibility ratings. ANOVA analyses revealed statistically significant differences between the word accuracy and intelligibility scores for the three groups.