Automated Snoring Detection System Using Synchrosqueezed Wavelet Transform

This paper presents a new approach for the recognition of snoring audio signals by applying advanced signal processing method, synchrosqueezed wavelet transform (SWT) and various machine learning classifiers such as VGG19, ResNet50, MobileNetV3, MobileNetV1, EfficientNetB7, DenseNet121, Xception, MobileNetV2, GoogleNet, and InceptionV3. Initially, the audio signals are preprocessed and further time-frequency representations (TFRs) have been obtained with the help of SWT. The obtained TFRs are then inputted into the machine learning classifiers to classifiy the snoring and non-snoring audio signals. Performance metrics in terms of accuracy, specificity, sensitivity, F1-score, and Matthews correlation coefficient (MCC) are computed. Highest accuracy of 96.5% is obtained by using the classifiers, VGG19 and ResNet50. This proposed methodology is suitable for real-time detection of snoring and can be used for diagnosis of sleep disorders such as sleep apnea.