Automated measurement of expressive prosody in neurodevelopmental disorders

Background: Autism spectrum disorders (ASD) are often associated with impaired expressive prosody. Existing methods for evaluating prosodic performance rely on time-consuming, subjective human judgments rather than automated methods. Objectives: The purpose of the study was to establish the validity of our automated digital measures of prosody by comparing those measures with human judgments. Methods: Responses for the following four tasks were scored using automated methods: (i) Lexical Stress (repeat a bisyllabic word with initial or final stress). (ii) Phrasing (describe a picture, indicating the number of pictured items, e.g., “chocolate, cookies, and jam” vs. “chocolate-cookies and jam”; adapted from PEPS-C (Peppé & McCann 2003)). (iii) Pragmatic Style (talk about a picture using prosody appropriate to address a baby vs. an adult). (iv) Focus (correct an inaccurate description of a picture, e.g., if the recording describes a blue cow as green, the subject responds “BLUE cow”; adapted from PEPS-C). The methods capture features of the melodic and temporal patterns that distinguish between contrasting speech responses. Four judges listened to forty utterance pairs for each task, with each pair from the same speaker with the same content but different target prosody. The judges were to identify the intended meaning of the two utterances (e.g., of two recordings, which one was meant to be “BLUE cow” rather than “blue COW”). Results: For all four tasks, the objective measure correlated with the weighted mean rating as well as or better than some of the judges correlated with each other. Conclusions: The automated digital measures were shown to be comparable in reliability to subjective judgment. Using these time-saving automated measures may eventually differentiate between the diagnostic groups more accurately for some tasks than human judgment-based measures and potentially lead to new speech-based markers for ASD.