This paper focuses on designing a robust stereophonic measurement set-up for sound sleep recording. The system is employed throughout the night during polysomnographic examinations of children in a pediatric sleep laboratory at a university hospital. Deep learning methods were used to classify the sounds in the recordings into four categories (snoring, breathing, silence, and other sounds). Specifically, a recurrent neural network with two long short-term memory layers was employed for classification. The network was trained using a dataset containing 1500 sounds from each category. The deep neural network achieved an accuracy of 91.16%. We developed an innovative algorithm for sound classification, which was optimized for accuracy. The results were presented in a detailed report, which included graphical representations and sound categorization throughout the night.
Keywords: classification of sleep sounds; deep learning; neural networks; polysomnography; sleep sounds; stereophony.