Cry detection can be defined as a procedure where the voiced crying sounds are selected from the recording. The most difficult part of the cry detection is to recognize the inspiratory sounds and separate them from the voiced sounds. In addition, sound recordings may come from different places and recorded with several devices, in this way the method of the cry detection has to be universal. The authors created the Extended Harmonic Product Spectrum method to classify the spectral structure of a given signal. Based on this new method the authors developed the Automatic Infant Cry Detection (AICD) system to detect voiced cry sounds in any kind of recording.

Index Terms. infant cry, cry detection, extended harmonic product spectrum