Application development for recognizing type of infant's cry sound

Abstract

Crying infant is a sign of baby who has a problem. But, some people are not able to recognize the meaning of infant's cry. Several researches to recognize infant's cry sound had been done by some researchers, but there is still no research that develop an application which able to recognize type of infant's cry sound based on web. In this research, an application is developed to help users identify the sound of crying infant based on Dunstan Baby Language. The method applied in this application are Mel-Frequency Cepstral Coefficient (MFCC) feature extraction for infant's cry sound, normalization of feature extraction result, and K-nearest neighbor classification. From the various tests performed, it can be concluded that highest average accuracy of 75.95 percent can be obtained by using parameters consist of 0.08 seconds wintime in MFCC feature extraction, 85 percent of training data and 15 percent of test data from any type of infant's cry sound, feature extraction normalization by Standard Deviation Normalization, and K-nearest Neighbor with k equal to 1 classification. While testing application by using all data, average accuracy of 96.57 percent can be obtained by using parameters consist of 0.08 seconds wintime in MFCC feature extraction, 85 percent of training data from any type of infant's cry sound, feature extraction normalization by Standard Deviation Normalization, and K-nearest Neighbor k equal to 1 classification. From that test, it can be concluded that the application has been running well when classifying all types of infant's cry sound data.