Exploring an Explanatory Child Speech Intelligibility Model Using Phonetically Contrasted Word Productions

Abstract

Purpose: In clinical practice, there is no standard measure of intelligibility that explains the kinds of difficulties that listeners who are inexperienced with child speech may encounter. This study aimed to investigate phonetic-based error types that occur in the speech of young children that contribute to decreased intelligibility. Method: Speech recordings of 9 preschool children producing phonetic contrasts that reflect common phonological disorders were analyzed by inexperienced listeners. To investigate the type of difficulties listeners in the general population might encounter, participants were recruited through the crowdsourcing platform Amazon Mechanical Turk (AMT). Testing the effect of phonetically contrasted words on the listeners’ ability to recognize the word and rate the intelligibility of the word was housed through a web-based platform compatible with AMT, Intelli-turk©. It was hypothesized that listeners inexperienced with child speech could rate the speech of children using Direct Magnitude Estimation and reflect different levels of intelligibility in agreement with previous word production accuracy measures. Results: The results of this study support the correlation between measures of whole-word accuracy and ratings of intelligibility. It was also found that different types of errors may contribute to listeners’ intelligibility ratings, meaning that listeners inexperienced with child speech productions identify differences in intelligibility categorically. Specific types of errors that contribute to the confusion of listeners’ intelligibility rating differed according to speaker accuracy level and phonetic contrast categories. This preliminary investigation yielded promising results towards establishing an explanatory model of intelligibility for preschool age children.