Automatic segmentation and pitch/jitter tracking of sleep disturbed breathing sounds

Obstructive sleep apnoea (OSA) is a serious disease disorder affecting a large percentage of the population. The current standard of OSA diagnosis known as polysomnography (PSG) is inconvenient and expensive, making it unsuited for community screening. Recently, a few researchers reported using snore related sounds (SRS) to diagnose OSA. Diagnostic systems intent on using SRS faces the tough problem of how to define an SRS sound and automatically segment into various components found there, in this paper, we present working definitions of SRS sounds. We propose algorithms to segment and categories them into pure breathing, silence, voiced/unvoiced snoring sounds and speech. Voiced snoring sound segments were then further analysed for the parameter pitch-jitter. This allowed us to track the development of OSA events starting from the time a subject goes to sleep. Results we obtained from real-world snoring sounds indicate that snore-segmentation followed by pitch-jitter estimation provides an efficient strategy to track the development of OSA.