Takahiro Emoto, Udantha R. Abeyratne, Yusuke Aoki, Masatake Akutagawa, Eiji Kondo, Ikuji Kawata, Shinsuke Konaka, and Yohsuke Kinouchi
snore related sound detection, neural network, connection weight space analysis
Snoring is the most common and characteristic symptom of the Obstructive Sleep Apnea(OSA) .The snore sound has been recently recorded during sleep for the purpose of OSA screening. The snore-related sound (SRS) as well as the silence is included in the recorded sound. The SRS detection plays an important role as a first step in snore segmentation. However, the SRS is complex signal, and at both high and low signals to noize ratio (SNR). And thus, the complexity and low SNR of signals make it a challenging task to detect them in the recorded sound. In this paper, we propose the novel method to detect automatically SRS by using the noise-robust neural network technique. The performance of the proposed method is evaluated on the clinical SRS data and compared with that of conventional zero-crossing-based method. We show that the proposed method can detect accurately the SRS compared to the conventional method. Even at very low SNR, the proposed method works within the detection error of 0.12[s].
Important Links:
Go Back
