Hybrid Machine Learning and Deep Learning-Based Infant Cry Classification for Automated Need Detection

The fact that infant needs can be interpreted using their crying patterns poses a central challenge to the early childcare practice in that infants rely on their cries as their primary form of communication. The paper illustrates a need detection system that is automated and involves machine learning and deep learning to categorize infant cries. The system takes in audio recordings of infant cries and categorizes them based on the classes of hunger and pain and discomfort and fatigue. The preprocessing stage of the audio signal processing is the application of data augmentation methods that entail the addition of noise to the audio signal and pitch shifts to strengthen the audio signal. The system isolates an entire set of acoustic features that comprise of MFCC and Mel Spectrogram and Chroma and Spectral Contrast and Zero Crossing Rate to quantify both the temporal and frequency characteristics. The classification model is a stacking-based ensemble of machine learning models (Random Forest, Support Vector Machine, K-Nearest Neighbors, and Gradient Boosting) in addition to a Convolutional Neural Network trained on spectrogram images. The system uses a weighted fusion process to fuse the prediction results of the two models. The experimental findings indicate that the proposed system has an overall accuracy of 93.8 per cent that is better than that of the individual models. It is a web-based application that runs on Flask allowing customers to make real-time predictions. The study introduces a smart healthcare solution that provides an efficient and scalable analysis of infant cries via the developed system.