| Abstrakt: |
Due to improper lifestyle, sleep disorders are becoming increasingly common worldwide. Early detection may help in preventing diseases arising due to sleep disorders, such as insomnia, muscle loss, breathing, and cardiac disorders. In this paper, nocturnal human sounds are analysed to develop a personal sleep monitoring system. Multiple audio-related features are extracted from the spectrograms of sleep sounds and analysed for discriminatory ability. The selected features are given as input to a fully-connected Artificial Neural Network (ANN) to classify the sleep sounds. The proposed approach classifies the considered seven categories of sleep sounds, including coughing, laughing, screaming, sneezing, snoring, sniffling, and farting, with an average accuracy of 97.4%. This is significantly higher than the classification accuracy obtained by applying conventional machine learning models on the selected features. This indicates that the ANN learns new features to enhance the classification accuracy of the sleep sounds. Moreover, the computational requirement of the system is kept low by reducing the number of features given as input to the ANN classifier. The proposed approach may be integrated with a smartphone or a cloud platform to develop a device for sleep monitoring or diagnosis of sleep disorders. [ABSTRACT FROM AUTHOR] |
Full text from SpringerLink