Automated Detection and Classification of Newborn Crying with Machine Learning.
Serap Özdemi̇r1, Efe Çetin Yilmaz2
1Gaziantep University, Faculty of Health Sciences, Department of Pediatric Nursing, Gaziantep, Türkiye.
Journal of Voice : Official Journal of the Voice Foundation
|November 18, 2025
Summary
Machine learning (ML) accurately predicts newborn needs from crying sounds, classifying behaviors like hunger and pain. This contactless, inexpensive system aids healthcare professionals in understanding infant cues.
Area of Science:
- Neonatal care
- Machine learning applications
- Bioacoustics
Background:
- Understanding newborn communication is crucial for timely care.
- Crying is a primary method for infants to express needs.
- Objective analysis of infant cries can be challenging.
Purpose of the Study:
- To develop a machine learning (ML) model for predicting newborn needs based on crying behavior.
- To classify infant cries into six basic behavioral patterns.
- To support perceptual analysis of newborn cries using an ML approach.
Main Methods:
- Real-time audio data collected from 32 newborns over 12-hour periods.
- Audio signals transformed into spectrograms using short-time Fourier transform.
- Deep convolutional neural networks and support vector machines used for classification.
Main Results:
- Accurate classification of six distinct newborn behavioral patterns (resting, hungry, sleepy, pain, burp, distress).
- Achieved approximately 91.856% similarity in cry unit band time analysis.
- Developed a contactless and inexpensive system suitable for clinical use.
Conclusions:
- The ML system effectively predicts and classifies newborn crying behaviors and needs.
- This technology can assist healthcare professionals in providing timely and appropriate care.
- The system's accuracy and cost-effectiveness make it valuable for routine clinical practice.
