Automatic snoring sounds detection from sleep sounds based on deep learning.

Area of Science:

• Biomedical Engineering
• Signal Processing
• Artificial Intelligence

Background:

• Snoring is a key indicator of obstructive sleep apnea hypopnea syndrome (OSAHS).
• Accurate snoring detection is crucial for OSAHS diagnosis and management.
• Automated methods can improve the efficiency and accessibility of sleep monitoring.

Purpose of the Study:

• To develop an automatic snoring detection algorithm for classifying snore and non-snore sound segments.
• To evaluate the effectiveness of different audio features and deep learning models for snoring detection.
• To assess the potential for using this algorithm in portable sleep monitoring devices.

Main Methods:

• Sleep sound signals were segmented using spectral entropy.
• Audio maps including time-domain waveform, spectrum, spectrogram, Mel-spectrogram, and CQT-spectrogram were generated.
• Two deep learning classifiers, CNNs-DNNs and CNNs-LSTMs-DNNs, were employed for classification.
• Convolutional Neural Networks (CNNs), Deep Neural Networks (DNNs), and Long Short-Term Memory (LSTM) networks were utilized.

Main Results:

• Mel-spectrograms provided better differentiation between snore and non-snore segments compared to other audio maps.
• The CNNs-LSTMs-DNNs model, utilizing Mel-spectrogram features, demonstrated strong performance for snoring detection.
• The developed algorithm shows promise for integration into portable sleep monitoring systems.

Conclusions:

• The Mel-spectrogram is a highly effective feature for snoring detection.
• Deep learning models, particularly CNNs-LSTMs-DNNs, are well-suited for analyzing audio features of snoring.
• The proposed method offers a viable solution for automated, portable sleep apnea monitoring.