Automated Classification of Snoring and Non-Snoring Sound Signals Based on Singular Superlet Transform

This paper proposes a new method for automated classification of snoring and non-snoring sound signals. The snoring and non-snoring sound signals have been represented using time-frequency representations (TFRs) obtained from singular superlet transform (SST). The obtained TFRs have been given as an input to various machine learning based classifiers, like convolutional neural network (CNN), visual geometry group 16-layer network (VGG-16), EfficientNet-B0, and hybrid VGG-16 models for the classification of snoring and non-snoring sound signals. From the experimental study, it has been observed that the hybrid VGG-16 with support vector machine (SVM) model has provided best performance among all the studied models with an accuracy, specificity, sensitivity, F1-score, and Matthews correlation coefficient (MCC) of 96.00%, 95.60%, 96.40%, 0.9602, 95.63%, and 0.9200 respectively. The good performance measures in automated classification of snoring and non-snoring sound signals show the effectiveness of the proposed method. The proposed method finds applications in the area of sleep disorder diagnosis and it is suitable for real-time implementation.