The role of artificial intelligence in detecting babies crying nature

Artificial intelligence has a very important role in medicine especially when huge amount of dataexists with insufficient knowledge. By analyzing the existing data with appropriate applicationsmany complex health problems can be solved.One of the important issues in pediatrics is the ability to understand the meaning of the babies cry.The characteristics of the cry are different in healthy from unhealthy infants and even is differentin various diseases. Also infants show different type of crying in situations such as pain, hunger,birth, and pleasure which can be recognized by trained listeners with the accuracy of ۳۳.۰۹%while the accuracy can be increased to ۸۰.۵۶% with artificial intelligence. The recognition of thecrying type could help physicians as well as disappointed parents and opens the way for productionof robot caregivers.To analyze the infant cry five stages including data acquisition, pre-processing, feature extraction,feature selection, and classification should be considered.For data acquisition the infant cry sounds should be recorded and labeled. In pre-processing stage,the background sounds should be removed and audio segmentation using Voice Activity Detectionshould be done. Feature extraction is an important stage; Discriminative features from theaudio signals in time or frequency domains should be extracted by computers and analyzed usingMel-frequency cepstral coefficient (MFCC), Linear Prediction Cepstral Coefficients (LPCC) andBark Frequency Cepstral Coefficients (BFCC), prosodic information (Variations in intensity, fundamentalfrequency, formants and duration) and Spectrograms. Then subgroups of features areselected in feature selection stage. Next the cry signals are classified to normal types of crying(hunger, sleepiness…) or pathologic ones such as asphyxia, hypo-acoustic (hearing disorder),cleft palate, respiratory distress syndrome, autism, etc.We are currently building a large infant cry database consisting of cries of infants from ۰ to ۹months old using the above mentioned method and we hope to be able to gather over ۳۰,۰۰۰ samplesreaching ۵۰ h of recording, which fits the need of deep neural networks.We are interested in creating a large database, extracting more strong and accurate features, establishingnew neural networks by using artificial intelligence for Iranian babies. We hope that withthis study, we will be able to quickly diagnose the baby’s problem by hearing his cry.

کلیدواژه ها:

artificial intelligence ، baby cry ، pediatrics