AUTOMATIC CLASSIFICATION OF INFANTS' HUNGER AND PAIN CRY.

Infants' cries convey information concerning pathologic and nonpathologic states. An automatic classification system has been developed for diagnosing various types of infants' cry. The system is based on the acquisition of cry features and its comparison via a distance measure to an a priori known template. The system was checked on two well-defined types of cry: hunger and specific pain cry. Classification tests performed on the initial database show typical error rates of less than 5 percent. The system will be tested for use as a diagnostic assist device for early detection of abnormalities such as hyperbilirubinemia or brain damage.