Infant Crying Detection and Classification Using Deep Learning

Abstract

Infant cry are crucial for various reasons. Cry pattern can be indicative of underlying medical issue such as belly-pain, discomfort, burping, hunger, tired and many more.  By recognizing and responding different cry patterns caregivers can also faster a stronger bond with their baby. This technology can be particularly beneficial for parent with disability insuring their baby’s needs are met even they cannot response immediately. Infants communicate via their cries. This research aims for classify the cry signals with using MAMBA state space model  and Audio Spectrogram Transformer model with the help of Mel-Spectrogram images. The primary objective of this to compare the effectiveness model such as MAMBA and AST model. The proposed system utilizes a dataset of labeled audio signals from infants. We use a Donate-a-cry corpus dataset. The image based features through the spectrogram which are extracted using pre-trained model.