Automatic snoring sounds detection from sleep sounds via multi-features analysis.

Area of Science:

• Biomedical Engineering
• Sleep Medicine
• Signal Processing

Background:

• Obstructive sleep apnea hypopnea syndrome (OSAHS) is a significant respiratory disorder.
• Snoring is a primary symptom of OSAHS, making snore analysis crucial for diagnosis.
• Automatic segmentation of snoring sounds is a critical preliminary step for snore analysis technology.

Purpose of the Study:

• To develop an automatic snoring detection algorithm for OSAHS diagnosis.
• To evaluate the effectiveness of linear and nonlinear features for snore classification.
• To compare the performance of spectral entropy (SE) and sample entropy (SampEn) in snore detection.

Main Methods:

• An adaptive effective-value threshold method was used to detect potential snoring episodes.
• Linear and nonlinear features, including maximum power ratio, sum of positive/negative amplitudes, 500 Hz power ratio, SE, and SampEn, were extracted.
• A support vector machine (SVM) was employed for automatic snore/nonsnore classification.

Main Results:

• Sample entropy (SampEn) demonstrated higher classification accuracy compared to spectral entropy (SE).
• The proposed automatic detection method achieved an accuracy exceeding 94.0% in classifying snoring and nonsnoring sounds.
• Effective classification was achieved even with small training datasets.

Conclusions:

• The developed automatic snoring detection method effectively classifies snoring and nonsnoring sounds.
• The algorithm's high accuracy and sensitivity support its utility in enabling automatic snoring detection for OSAHS diagnosis.
• SampEn is a promising feature for improving the accuracy of snoring detection algorithms.