Searching Robust Acoustical Indicators for Hypoxia-Related Disorders in Neonates for Classification with Neural Networks

For many years, research efforts have been made to investigate whether the sound of a crying infant can be analyzed to determine various aspects concerning the infant’s health. In recent publications, possible acoustical indicators for several health disorders have been proposed. The goal of this thesis assignment is to investigate several combinations of these acoustical indicators in order to find a robust set of parameters for discriminating between normal (healthy) and abnormal (suffering from a hypoxia-related disorder) cases. These parameters will first be extracted from the crying sound, after which attempts will be made to automatically classify infant cries as either normal or abnormal, based on these parameters. This classification will be accomplished using neural networks especially suited for classification of feature patterns.