Machine learning-based infant crying interpretation.

Area of Science:

• Infant communication and signal processing.
• Machine learning applications in healthcare.

Background:

• Infant crying is a primary communication method, often difficult for caregivers to interpret.
• Misinterpreting infant cries can lead to significant problems for infant well-being.
• Existing methods for cry analysis require improved audio feature representation and classification.

Purpose of the Study:

• To develop an accurate infant cry analysis system.
• To evaluate various audio feature representations and machine learning classifiers for cry interpretation.
• To identify optimal features and models for distinguishing infant states based on vocalizations.

Main Methods:

• Utilized time-domain (Zero-Crossing Rate, Root Mean Square), frequency-domain (Mel-spectrogram), and time-frequency-domain (Mel-frequency cepstral Coefficients - MFCCs) audio features.
• Applied time-series imaging algorithms to transform MFCC features into visual representations.
• Trained and evaluated multiple machine learning classifiers including Decision Tree, Random Forest, K-Nearest Neighbors, and Bagging.

Main Results:

• Mel-frequency cepstral Coefficients (MFCCs), Zero-Crossing Rate (ZCR), and Root Mean Square (RMS) features demonstrated high performance.
• The MFCC-based Random Forest (RF) classifier achieved a 96.39% accuracy rate.
• This performance surpassed the state-of-the-art scalogram-based ShuffleNet classifier (95.17% accuracy).

Conclusions:

• The proposed MFCC-based Random Forest approach offers a highly effective method for infant cry analysis.
• This study highlights the importance of advanced audio feature extraction and machine learning for understanding infant communication.
• The findings provide a promising tool to aid caregivers in interpreting infant needs more accurately.