Comparison of Supervised-Learning Models and Auditory Discrimination of Infant Cries for the Early Detection of Developmental Disorders / Vergleich von Supervised-Learning Klassifikationsmodellen und menschlicher auditiver Diskriminationsfähigkeit zur Unterscheidung von Säuglingsschreien mit kongenitalen Entwicklungsstörungen

Abstract

Infant cry classification can be performed in two ways: computational classification of cries or auditory discrimination by human listeners. This article compares both approaches. An auditory listening experiment was performed to examine if various listener groups (naive listeners, parents, nurses/midwives and therapists) were able to distinguish auditorily between healthy and pathological cries as well as to differentiate various pathologies from each other. Listeners were trained in hearing cries of healthy infants and cries of infants suffering from cleft-lip-and-palate, hearing impairment, laryngomalacia, asphyxia and brain damage. After training, a listening experiment was performed by allocating 18 infant cries to the cry groups. Multiple supervised-learning classifications models were calculated on the base of the cries’ acoustic properties. The accuracy of the models was compared to the accuracy of the human listeners. With a Kappa value of 0.491, listeners allocated the cries to the healthy and the five pathological groups with moderate performance. With a sensitivity of 0.64 and a specificity of 0.89, listeners were able to identify that a cry is a pathological one with higher confidence than separating between the single pathologies. Generalized linear mixed models found no significant differences between the classification accuracy of the listener groups. Significant differences between the pathological cry types were found. Supervised-learning classification models performed significantly better than the human listeners in classifying infant cries. The models reached an overall Kappa value of up to 0.837.