Spectrogram-based fundamental frequency tracking of spontaneous cries in preterm newborns

Cry analysis of preterm newborns has proven to be relevant for prediction of pathologies or for comparison with full-term newborns. In this paper we propose a new approach for the automated detection and tracking of the fundamental frequency in cries, based on the processing of the spectrogram. A first step automatically detects the frequency bounds including the fundamental frequency along each cry. Then, the tracking of the fundamental frequency is obtained after a contour detection. Results showed that this new approach allows to process efficiently all types of cries. This whole procedure applied to a database including 1889 cries from 14 babies, at term-equivalent age, highlighted differences between extremely, very and late preterm as well as full-term newborns. In addition, we observed a decrease of the mean fundamental frequency with increasing gestational age, a result in accordance with the literature.