A data-driven approach for perceptually validated acoustic features for children's sibilant fricative productions

Both perceptual and acoustic studies of children's speech independently suggest that phonological contrasts are continuously refined during acquisition. This paper considers two traditional acoustic features for the 's'-vs.-'sh' contrast (centroid and peak frequencies) and a novel feature learned from data, evaluating these features relative to perceptual ratings of children's productions. Productions of sibilant fricatives were elicited from 16 adults and 69 preschool children. A second group of adults rated the children's productions on a visual analog scale (VAS). Each production was rated by multiple listeners; mean VAS score for each production was used as its perceptual goodness rating. For each production from the repetition task, a psychoacoustic spectrum was estimated by passing it through a filter bank that modeled the auditory periphery. From these spectra centroid and peak frequencies were computed, two traditional features for a sibilant fricative's place of articulation. A novel acoustic measure was derived by inputting the spectra to a graph-based dimensionality-reduction algorithm. Simple regression analyses indicated that a greater amount of variance in the VAS scores was explained by the novel feature (adjusted R² = 0:569) than by either centroid (adjusted R² = 0:468) or peak frequency (adjusted R² = 0:254).