On the use of hidden Markov models in infants' cry classification

Since the beginning of the 1960's efforts have been made to characterize pathologies in infants' cries. In various pathologies such as brain damage, cleft palate, hydrocephalus, sudden infant death syndrome (SIDS), and many other diseases, special cry characteristics have been reported. The studies performed on the subject (most employing manual time and/or frequency analysis) established the basis for the development of models for cry production and also for the beginning of the development of automatic classification systems. Such a system has been suggested by Cohen and Zmora (Proc. Int. Conf. Dig. SIGPRO, Cappellini, V. and Constantinides, A.G., Eds., Elsevier, Amsterdam, 1984). The system has been proven to be successful in classification of a preliminary database which consisted of hunger and pain cries of healthy full-term infants. Based on this previous work, this paper introduces an automatic classification system for diagnosing various types of infants' cries. The system is based on continuous density hidden Markov models (CD-HMM).