Infant Cry Classification Using Deep Learning and Feature Fusion

A infant's cry is their primary form for interacting and can provide essential information about their wants or condition. In the current research, we describe an innovative approach for classifying baby sounds using the combination of features and deep learning strategies. The proposed method integrates criteria including spectral contrast, chroma, and Mel-Frequency Cepstral Coefficients (MFCC) to gather many different kinds of infant shouts' sounds. We use a hybrid model that combines Recurrent Neural Networks (RNNs) for temporal sequence analysis and convolutional neural networks (CNNs) for feature extraction. Experiments show how well the suggested design works resulting in useful precision as well as stability throughout various cry groups. The results suggest this kind of technology may be able to be used in real-time in parenting and health care environments. For more accurate cry level prediction, the CNN gathers spatial information and the RNN tracks temporal relations. A classified infant cry dataset is used for verifying the approach, and metrics which involves precision, recall, accuracy, and F1-score are analyzed is to measure results. By contributing to develop technology for baby care, the study improves parental ability for satisfying the needs of the infant.