The use of functional data analysis to study variability in children’s speech

Abstract

A common finding in studies of children’s speech is increased variability relative to adults. This result may be observed well after the age when basic phonological contrasts are established. Researchers debate the sources of this variability, but most agree that the study of variability, and its decrease over time, may provide insight into motor control development. Recent data on oral airflow during /h/ in 5-year-olds and adults revealed that the children presented greater relative amplitude variability than the adults, and seemed to show less regularity in the rates of ad/abduction for /h/. This presentation applies functional data analysis (FDA) to the airflow data from two women and one 5-year-old speaker. In FDA, signals are nonlinearly normalized using a smooth warping function of time, and separate indices are obtained for phasing (temporal) and amplitude variability. Results indicate that the child shows greater variability in both temporal and amplitude measures, but the extent of adult–child differences depends somewhat on the specific methods of normalization and alignment. FDA shows potential as a means of gaining more detailed information about the nature of variability in children’s speech data. [Work supported by NIH and CNPq, Brazil.]