Newborn´s pathological cry identification system

In this paper we compare the performance of an identification system of the pathological and normal cries of the newborn, using various methods of characterisation of cries. This system is similar to a speaker identification system. It contains two main parts namely a cry signal characterisation and modeling. We used Mel-Frequency Cestrum Coefficients and Mel Frequency Discret Wavelet Coefficients to characterize the newborn cry signals. We also applied Best Structure Abstract Tree algorithm and the Principal Component Analysis to reduce the number of Wavelet packet transform WPT coefficients. In this study a Probabilistic Neural Network classifier is used. The best result obtained is 96.99% of correct identification using Best Structure Abstract Tree algorithm.