15th International Congress of Phonetic Sciences (ICPhS-15)

Barcelona, Spain
August 3-9, 2003

Working in a Robust Non-Linear Infant Cry Classifier

Sergio D. Cano Ortiz (1), Daniel I. Escobedo Beceiro (2), Taco Ekkel (1)

(1) University of Oriente, Cuba
(2) University of Twente, The Netherlands

The paper pretends to implement a parametric classification to lead to a robust parameter set which permit to find out the most efficient cry parameters for a robust classification of the cry units. It's achieved through an intelligent searching algorithm combined with a fast non-linear classification procedure, establishing the cry parameters which better matches the physiological status previously defined for the six control groups used as input data. Finally the optimal acoustic parameter set is chosen in order to implement a new non-linear classifier based on a radial basis function network, a neural NN-based procedure which classifies the cry units into a 2 categories, normal-or abnormal case. This study also represent one of the few attempts at the application of Artificial Neural Networks (ANN's) in the domain of infant cry as well as the first one oriented for diagnoses purpose on hypoxaemia-based diseases, at least on the documented results in the literature.

Full Paper

Bibliographic reference.  Cano Ortiz, Sergio D. / Escobedo Beceiro, Daniel I. / Ekkel, Taco (2003): "Working in a robust non-linear infant cry classifier", In ICPhS-15, 3077-3080.