Comparative Analysis of Infant Cry Classification using Machine Learning and Deep Learning Techniques

Infant Cry Classification is the process of analyzing acoustic signals from an infant's cry for extracting relevant features to determine the cause of crying, and assess the infant's health condition. The process is typically divided into four main stages: pre-processing, feature extraction, feature fusion, and classification. In the feature extraction stage, spectrograms along with frequency transforms such as the Wavelet Transform, Cosine Transform, and Fourier Transforms are employed to capture important characteristics of the cry. The feature fusion stage is critical as it integrates information from various scales and contexts obtained during feature extraction, allowing for a more comprehensive representation of the cry's features. In current times, both Machine Learning (ML) and Deep Learning (DL) techniques are employed for classification. The ML techniques predominantly utilized in this process are: Multi-Class Support Vector Machines (SVM), Hyperparameter-Optimized SVM, Kernel-based SVM, Ensemble Boosting methods, Random Forest classifiers, K-Nearest Neighbors (K-NN), and Bayesian Optimization. In the case of DL architectures, models such as Style and Latent Guided Adversarial Networks (SLGAN), Multi-Layer Perceptron (MLP), Convolutional Neural Network - Sparse Convolution Net (CNN-SCNet), 1D CNN, and Dual-CNN are frequently employed to ensure high accuracy. The performance of these classification algorithms is evaluated in terms of the following metrics: sensitivity, specificity, accuracy, precision, and F-measure.