A Machine Learning Approach to Classify Biomedical Acoustic Features for Baby Cries

Area of Science:

• Infant communication
• Bioacoustics
• Machine learning in healthcare

Background:

• Newborns primarily communicate needs via crying.
• Understanding infant cries is crucial for timely care, posing a challenge for parents.
• The Dustan Baby Language categorizes cries into five basic needs: hunger, burping, fatigue, cramps, and discomfort.

Purpose of the Study:

• To develop an automated model for recognizing and classifying infant cries.
• To differentiate between infant cries indicating hunger versus discomfort or pain.
• To evaluate the efficacy of various machine learning models for this classification task.

Main Methods:

• Extraction of spectral features (chroma_stft, spectral_centroid, etc.) from infant cry audio.
• Implementation and comparison of four classification algorithms: Random Forest, Support Vector Machine (SVM), Logistic Regression, and Decision Tree.
• Training and testing models on labeled infant cry data.

Main Results:

• The Support Vector Machine model demonstrated superior performance in accurately classifying infant cries compared to other tested models.
• Spectral features extracted from cry signals were effective inputs for the classification models.
• The study successfully differentiated between cries related to hunger and those indicating discomfort.

Conclusions:

• Machine learning, particularly SVM, can effectively interpret infant cries to identify specific needs like hunger or discomfort.
• This technology has the potential to aid caregivers in responding more promptly and accurately to infant distress.
• Further research can refine these models for broader applications in infant care and health monitoring.