Sound based snore signal processing

• Main Author: Ang, Shouchen
• Other Authors: Ser Wee
• Format: Final Year Project
• Language: English
• Published: 2013
• Subjects: DRNTU::Engineering
• Online Access: http://hdl.handle.net/10356/54242
• Institution: Nanyang Technological University

Snoring can be commonly found in up to 40% of the adult population [1] and it is a symptom that can be used as an indication to justify the existence of obstructive sleep apnoea syndrome(OSA). Over the years, studies have proven that the acoustic characteristics of the snoring sound may differ between a simple/normal snorer and an OSA patient. Henceforth, this project aims to create an algorithm to evaluate the sound signal of a small group of simple snorers and OSA patients. It will analyse the acoustic characteristics of each of their snoring signal and make an identification on the possibility of a patient being diagnosed with OSA. The snoring sound signal of 10 patients were recorded during a full night polysomnography test done by National University Hospital(NUH). Upon completion of the test, certified medical doctors will evaluate the data recorded and conduct several more extensive test before finally diagnosing a patient with OSA. Thereafter, the snoring data along with their respective diagnosis were provided to the author for the usage in this project. From the spectral analysis of the snoring sounds, there appear to be an existence of two distinct pattern between a simple snorer and an OSA patient. For a simple snorer, the pattern is characterized by a fundamental frequency and distinct harmonics. However, for the latter, they generally have lower frequency peak, whereby the threshold is usually below 150Hz [2] and without any clear harmonics. Therefore with this knowledge, an algorithm was written and the peak frequency was used as a cut to identify the two different groups of snorers. Several evaluation has been done before concluding on a peak frequency threshold cut at 100Hz for this project. As long as the peak frequency of the snore signal falls below 100Hz, it will generate an indication to show the patient will potentially be diagnosed with OSA. With this threshold in place, it is found that there is an 84% success in rightfully identifying whether a patient has OSA or not. This is a 40% increase in succession from the usage of the threshold obtained from earlier research that resulted in a 44% accuracy. Henceforth, this report will discuss on the design process as well as the implementation of the algorithm. It will concluded by discussing on the results obtained and how future works can be implemented to further improve this project.