Infant Cry Classification using CNN-MFCC Fusion

Classifying infant cries is crucial to correctly identifying a newborn's needs or health concerns. This paper presents a novel approach that uses Mel-Frequency Cepstral Co-efficients (MFCCs) and convolutional neural networks (CNNs) to automatically classify baby cries. Effective cry signal en-coding is accomplished by utilizing MFCC features to capture distinctive acoustic qualities. CNN architecture is used to process these features, allowing for the accurate classification of baby cries into categories like hunger, discomfort, pain, and tiredness. The model's performance on a variety of datasets demonstrates its potential for early identification of health issues or specific needs in newborns. Furthermore,The mathematical outcomes of the conclusion indicate that the CNN model's accuracy was 94%, which was marginally lower than the 96% accuracy of the Decision Tree model.. In spite of this, the CNN is a better choice due to its strong generalization, ability to control class imbalance, and potential for overfitting. By merging CNN and MFCC, this method provides a dependable and adaptable means of automatically interpreting infant cries, advancing the field of technology-assisted healthcare applications for early childhood development.