Proposing Novel Recurrent Neural Network Architectures for Infant Cry Detection in Domestic Context

Infants must be monitored around the clock because they are at risk for various health issues and cannot communicate their needs properly. Though several infant monitoring systems exist to monitor infants, most of these did not explicitly focus on detecting infant cries, which is infants' primary form of communication. Automatic detection of an infant cry in a domestic context is essential in such systems. It is also crucial for researchers who study the relationship between infant cry patterns and various health parameters. This research proposes novel recurrent neural network architectures for automatically detecting infant cries in a domestic context that will achieve better performance and provide a more convenient way for parents to monitor their infants. To attain these objectives, four recurrent neural network architectures were built and trained with a dataset containing infants' recordings and various noises available in a domestic context. The models were also evaluated in terms of accuracy, precision, recall, and f1-score, As an outcome, the trained models were found to achieve more than 88 % on all the performance parameters, which proves the satisfactory performances of the developed models.