Table 1 Summary of deep learning algorithms for snoring detection.
From: Automatic snoring detection using a hybrid 1D–2D convolutional neural network
References | Year | Dataset/Environment | Equipment | Features | Classifiers | Results |
|---|---|---|---|---|---|---|
2015 | 15 subjects | A unidirectional mic. faced the neck | Raw audio signal | f-MLP | Accuracy: 94.8%-96.6% | |
2017 | 68 subjects Sleep Studies Laboratory | A mic. placed 30 cm from subject’s mouth | Frequency domain features | MLP | Accuracy: 81.2% | |
2018 | 20 subjects Hospital | A mic. placed 70 cm from the top end of the bed | MFCCs | RNN | Accuracy: 95% Sensitivity: 92% Specificity: 98% | |
2019 | 10 subjects Subject’s private bedroom | A phone placed 50 cm away from subject’s head | Raw audio signal | 1D CNN | Average precision: 81.82% | |
2019 | 1000 samples Online source | – | MFCCs | 2D CNN | Accuracy: 96% | |
2020 | 15 subjects Hospital | A mic. placed 45 cm above the subject’ s mouth | Mel-spectrogram | CNN–LSTM–DNN | Accuracy: 95.07% Sensitivity: 95.42% Specificity: 95.82% | |
2021 | 38 subjects Sleep Laboratory | A mic. placed 70–130 cm from the top end of the bed | CQT-spectrogram | CNN–LSTM | Accuracy: 95.3% Sensitivity: 92.2% Specificity: 97.7% |
