An investigation into audio features and DTW algorithms for infant cry classification

Cry is the most common phenomenon among infants, and it has been reported that babies cry for multiple reasons. Infant cry signals are thought to convey much useful information about the physiological and pathological state of the baby. Hence, in this work we analyzed these audio signals in order to classify different reasons of cries. Cry signals were especially collected for this study including three causes, namely hunger, pain and uncertainty. Modified MFCC features besides basic acoustic features were extracted from each recording. After intergroup variance examination, nine features were selected and subjected to a novel matching process based on Dynamic Time Warping (DTW) for separating infant cries. Experiment results show that nine selected features are effective to recognize cries caused by hunger, pain and other uncertain reasons. The proposed approach for infant cry analysis will provide useful information for designing towards an automatic system for detecting physiological and pathological state of the baby.