Optimizing Infant Cry Recognition: A Fusion of LPC and MFCC Features in Deep Learning Models

This study focuses on the recognition of infant cry sounds and aims to improve the accuracy and understanding of different cry types. The research endeavor remains ongoing, demanding sustained efforts to elucidate the root causes of newborn crying, discriminate between cry types in both normal and pathological contexts, and precisely distinguish conditions like hunger, pain, and stress. Furthermore, the intricate differentiation of various pathological conditions necessitates in-depth exploration. Through the application of a novel approach that combines feature extraction techniques and deep learning algorithms, the study achieves high levels of accuracy in classifying infant cries. The convolution neural network model emerges as the most effective algorithm, demonstrating robust performance in both balanced and imbalanced dataset scenarios. The integration of Linear-predictive coding (LPC) and Mel Frequency Cepstral Coefficients (MFCC) feature extraction methods allows for capturing comprehensive and discriminative features from infant cry audio signals. The methodology undergoes validation through extensive experimentation. The findings significantly contribute to the field of infant cry sound recognition by providing a robust methodology and emphasizing the importance of feature extraction and data representation. Overall, this study advances our understanding of infant cry sound recognition and carries practical implications for improving infant care and support.