Klasifikasi Tangisan Bayi Menggunakan Parameter Pitch Dengan K-Nearest Neighbors

Abstract

Baby crying is a basic and important thing for mothers or caregivers to understand. In general, young mothers who do not receive guidance from experienced people, usually interpret baby crying as a sign of hunger only, even though crying in babies has different meanings or types of crying depending on the trigger/cause of crying. This study was conducted to establish the characteristics of the cause of infant crying through pitch parameters formed in the Bag of Features and determine the accuracy of the resulting classification. The feature extraction and classification methods used in this research are pitch, Bag of Features and K-Nearest Neighbor. Pitch feature extraction is done by changing the range parameters and methods in estimating the fundamental frequency. The range and method used in this research are (70,170) and PEF. The baby cries used for this study were taken in two ways, namely downloading Dunstan Baby Language and field measurements based on the perception of mothers and medical personnel. The types of infant cries used in this study were burpme, hungry, lower wind pain, tired, uncomfortable and pain. The results of this study show that the sequence of DBL baby cry labels that have a high average fundamental frequency probability value based on the Bag of Features histogram are tired (0.290), lower wind pain (0.207), hungry (0.206), burpme (0.182) and uncomfortable (0.090) while the sequence of baby cry labels from measurement data shows that the sick label has a higher average fundamental frequency, which is 0.200 when compared to the hungry label whose average fundamental frequency is 0.064. The classification accuracy results obtained between the DBL database test and the measurement database using K-Nearest Neighbor look optimal, which is 92% and 98%.