A Pragmatic Approach for Infant Cry Analysis Using Support Vector Machine and Random Forest Classifiers

• Published in: Wireless personal communications Vol. 137; no. 4; pp. 2269 - 2280
• Main Authors: Basavaiah, Jagadeesh, Anthony, Audre Arlene
• Format: Journal Article
• Language: English
• Published: New York Springer US 01.08.2024; Springer Nature B.V
• Subjects: Babies; Classification; Communications Engineering; Computer Communication Networks; Discomfort; Engineering; Feature extraction; Infants; Machine learning; Networks; Pain; Signal,Image and Speech Processing; Support vector machines; Belly pain; Hunger; Discomfort; Infant cry; Classification; Random forest; Support vector machine; Burping
• ISSN: 0929-6212; 1572-834X
• DOI: 10.1007/s11277-024-11491-8

A baby’s first spoken communication comes through crying. Before learning to convey their psychological/physiological needs or feelings using language, babies typically express how they feel by crying. Crying is a reaction to an inducement like pain, discomfort or hunger. Nevertheless, it is difficult sometimes to understand why a baby is crying. This will be annoying for a parents/guardian/caretaker, and therefore in this work, we are proposing an infant cry analysis and classification system to classify the kinds of crying of babies to assist parents/guardian/caretaker and attend to the needs of the babies. Presently, five distinct kinds of infant cries are identified: hunger (Neh), belly pain (Eairh), tiredness (Owh), discomfort (Heh) and burping (Eh). The database of this study consists of 456 audio recordings of 7 s each of 0–22-week-old babies. Feature extraction from each crying frame is carried out using Mel-frequency cepstral coefficients and the sequential forward floating selection is later used to choose highly discriminative features. Support Vector Machine and Random Forest classifiers are used for classification of infant crying. The results of the experiments has shown the performance of the proposed system with a accuracy of classification of 78% and 90.8% for Support Vector Machine and Random forest classifiers respectively.