Infant cry analysis of cry signal segments towards identifying the cry-cause factors

This is the authors' abstract. We don't add key points for this paper.

Infants communicate their physiological state and emotions mostly by crying. Identifying the cause of crying is natural and easy for human beings, but challenging for machines. Automatic identification of cry-cause factors has vast applications in assistive healthcare and timely remedial measures in critical situations. In this paper, Infant cry signals analysis is carried out to identify four cry-causes, namely, Environmental Change, Discomfort, Stranger's Anxiety and Pain. Different causes of infant crying are characterized by examining changes in the acoustic features that are extracted from acoustic signals in the cry segments. Mainly, the instantaneous fundamental frequency (F 0 ) is extracted using auto-correlation of cry signal. Changes in F 0 contours for different cry-causes are examined by deriving parameters mean, standard deviation and normalized standard deviation. A Infant Cry Signals Database (IIITS ICSD 2) is used, which was especially collected for this study. Results are validated by observing changes in the F 0 and its harmonics in spectrograms, derived using short-time Fourier transform. Few distinct patterns of changes in the F 0 contours are observed for different cry causes. Pain cry is characterized by higher standard deviation in F 0 , than for Discomfort. The standard deviation in F 0 for Environmental Change is more than for Stranger's Anxiety. Normalized standard deviation is highest for Pain, than for Environmental Change, than for Discomfort and lowest for Stranger's Anxiety. These insights should be helpful towards automated identification of infant cry-causes and related numerous applications.