Brief Communication Could formant frequencies of snore signals be an alternative means for the diagnosis of obstructive sleep apnea? Andrew Keong Ng a, * , Tong San Koh a , Eugene Baey b , Teck Hock Lee a , Udantha Ranjith Abeyratne c , Kathiravelu Puvanendran d a School of Electrical and Electronic Engineering, Nanyang Technological University, Singapore 639798, Republic of Singapore b Respironics Incorporated, Vbox 881389, Singapore 919191, Republic of Singapore c School of Information Technology and Electrical Engineering, The University of Queensland, St. Lucia, Brisbane, Australia d Sleep Disorders Unit, Singapore General Hospital, Singapore 169608, Republic of Singapore Received 15 May 2007; accepted 18 July 2007 Available online 6 September 2007 Abstract Objective: To study the feasibility of using acoustic signatures in snore signals for the diagnosis of obstructive sleep apnea (OSA). Methods: Snoring sounds of 30 apneic snorers (24 males; 6 females; apnea–hypopnea index, AHI = 46.9 ± 25.7 events/h) and 10 benign snorers (6 males; 4 females; AHI = 4.6 ± 3.4 events/h) were captured in a sleep laboratory. The recorded snore signals were preprocessed to remove noise, and subsequently, modeled using a linear predictive coding (LPC) technique. Formant frequencies (F1, F2, and F3) were extracted from the LPC spectrum for analysis. The accuracy of this approach was assessed using receiver operating characteristic curves and notched box plots. The relationship between AHI and F1 was further explored via regression analysis. Results: Quantitative differences in formant frequencies between apneic and benign snores are found in same- or both-gender snor- ers. Apneic snores exhibit higher formant frequencies than benign snores, especially F1, which can be related to the pathology of OSA. This study yields a sensitivity of 88%, a specificity of 82%, and a threshold value of F1 = 470 Hz that best differentiate apneic snorers from benign snorers (both gender combined). Conclusion: Acoustic signatures in snore signals carry information for OSA diagnosis, and snore-based analysis might potentially be a non-invasive and inexpensive diagnostic approach for mass screening of OSA. Ó 2007 Elsevier B.V. All rights reserved. Keywords: Obstructive sleep apnea; Polysomnography; Snoring; Snore signals; Acoustic analysis; Formant frequencies; Linear predictive coding 1. Introduction Obstructive sleep apnea (OSA) is a common sleep- related breathing disorder, characterized by a cessation of respiratory for at least 10 s, corresponding to a com- plete upper airway (UA) occlusion despite continuous abdominal and chest wall movements. The gold stan- dard for diagnosing OSA is an overnight multi-channel polysomnography (PSG), which can be time-consuming and labour-intensive in setting up and in the subsequent analysis. Therefore, many researchers have attempted to search for other modalities, such as airflow [1], nasal pressure [2], and oxygen saturation [3], to detect OSA. However, these studies require at least a physical contact sensor, which may cause discomfort to the patients. In addition, specific expertise may be needed at the test site for correct placement of sensors. Snoring is a common and earliest symptom of OSA, affecting more than 80% of the OSA patients [4]. Approx- imately 87% of 220 habitual snorers studied in Singapore have OSA [5]. Snoring is caused by the vibration of soft 1389-9457/$ - see front matter Ó 2007 Elsevier B.V. All rights reserved. doi:10.1016/j.sleep.2007.07.010 * Corresponding author. Tel.: +65 94306840; fax: +65 67930756. E-mail address: ngke0002@ntu.edu.sg (A.K. Ng). www.elsevier.com/locate/sleep Sleep Medicine 9 (2008) 894–898