Automatic and Unsupervised Snore Sound Extraction From Respiratory Sound Signals

• Published in: IEEE transactions on biomedical engineering Vol. 58; no. 5; pp. 1156 - 1162
• Main Authors: Azarbarzin, Ali, Moussavi, Zahra M. K.
• Format: Journal Article
• Language: English
• Published: New York, NY IEEE 01.05.2011; Institute of Electrical and Electronics Engineers; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Acoustic noise; Airway management; Algorithms; Ambient recording; Applied sciences; Cardiovascular diseases; Cluster Analysis; Clustering algorithms; Detection algorithms; Exact sciences and technology; Feature extraction; Female; Fuzzy Logic; Hidden Markov models; Humans; Information, signal and communications theory; Male; Microphones; Middle Aged; Patient monitoring; Pattern Recognition, Automated - methods; Polysomnography - methods; Principal Component Analysis; Principal components analysis; Reproducibility of Results; respiratory sound analysis; Signal and communications theory; Signal Processing, Computer-Assisted; Signal representation. Spectral analysis; Signal, noise; Sleep apnea; Sleep Apnea Syndromes - physiopathology; snore sounds; Snoring - diagnosis; Snoring - physiopathology; Sound Spectrography - methods; Studies; Telecommunications and information theory; Trachea; tracheal recording; unsupervised clustering; tracheal recording; Acoustic signal; Ambient recording; Unsupervised classification; unsupervised clustering; Respiratory system; Signal classification; snore sounds; Respiratory tract; respiratory sound analysis; Audio signal; Sound analysis; Recording; Trachea; Biomedical engineering
• ISSN: 0018-9294; 1558-2531; 1558-2531
• DOI: 10.1109/TBME.2010.2061846

In this paper, an automatic and unsupervised snore detection algorithm is proposed. The respiratory sound signals of 30 patients with different levels of airway obstruction were recorded by two microphones: one placed over the trachea (the tracheal microphone), and the other was a freestanding microphone (the ambient microphone). All the recordings were done simultaneously with full-night polysomnography during sleep. The sound activity episodes were identified using the vertical box (V-Box) algorithm. The 500-Hz subband energy distribution and principal component analysis were used to extract discriminative features from sound episodes. An unsupervised fuzzy C-means clustering algorithm was then deployed to label the sound episodes as either snore or no-snore class, which could be breath sound, swallowing sound, or any other noise. The algorithm was evaluated using manual annotation of the sound signals. The overall accuracy of the proposed algorithm was found to be 98.6% for tracheal sounds recordings, and 93.1% for the sounds recorded by the ambient microphone.