System and method for automatic transcription of cry signals for analysis and to determine type and pathology
Publication Date: 2012-Mar-31
The IP.com Prior Art Database
Studies in the field of nueropsychology of infants have established links between infants' cries and their physical and psychological state. Cry analysis for disability detection can be a reliable, non-intrusive and cost-efficient solution for early detection of disabilities. We propose a fundamental approach by which a cry signal is transcribed as a sequence of cry units (much like the automatic transcription of speech signals into phones/words). We present our comparative analysis of spectral characteristics of healthy and pathological infants. Based on this analysis, we propose a class of spectral features to distinguish healthy infants' cries from that of pathological infants. Specifically, the proposed features utilize the difference in pitch-synchronous spectral entropy to distinguish the two cry types. The two-class classification accuracy of the proposed features is 9.2\% better than that obtained by the optimal settings for the standard MFCC. We also discuss various reasons that could lead to the observed differences in the characeteristics of the two types of cries.