Measures of variability in child and adult speech: Test–retest results

Abstract

Elevated measures of speech production variability are commonly observed in children and in clinical populations. Methods of assessing variability thus have utility in understanding processes of speech development and breakdown. Valid interpretation of variability measures for a given speaker requires understanding the stability of such measures over time. Here, we present test–retest data on aerodynamic measures of consonants in two 5-year-old children and two women, each recorded two times approximately 3.5 months apart. Two methods of assessing variability, the spatio-temporal index (STI) and Functional Data Analysis (FDA), were applied to the data. STI yields a single value over repeated productions, incorporating both spatial and temporal aspects of the data. FDA yields separate measures of warping (phasing) and amplitude. Both FDA and STI analyses yielded higher values in children than adults at both times, as expected. The test–retest analysis indicated that both STI and FDA values were significantly correlated over time, suggesting that either method provides a reliable measure of speech production variability. The r values for STI were higher than for FDA, whereas FDA values showed more differentiation across consonants. Our discussion focuses on the relative merits of these two approaches for assessing speech variability.