Selecting an acoustic correlate for automated measurement of American English rhotic production in children

AbstractPurpose: A current need in the field of speech–language pathology is the development of reliable and efficient techniques to evaluate accuracy of speech targets over the course of treatment. As acoustic measurement techniques improve, it should become possible to use automated scoring in lieu of ratings from a trained clinician in some contexts. This study asks which acoustic measures correspond most closely with expert ratings of children’s productions of American English /ɹ/ in an effort to develop an automated scoring algorithm for use in treatment targeting rhotics.Method: A series of ordinal mixed-effects regression models were fit over a large sample of children's productions of words containing /ɹ/ that had previously been rated by three trained clinicians. Akaike/Bayesian Information Criteria were used to select the best-fitting model.Result: Controlling for age, sex, and allophonic contextual differences, the measure that accounted for the most variance in speech rating was F3–F2 distance...