Tsuyoshi Mikami

Automatic classification of oral/nasal snoring sounds based on the acoustic properties

Snoring was once regarded as an indication of good sleep. But recently it has been known to be one of the symptoms which indicate sleep disordered breathing such as sleep apnea syndrome. Moreover, heavy snoring caused by oral breathing sometimes leads benign snorers to be apneics. Thus, it is important to detect oral snoring for medical treatment in the earlier stage, but we cannot know our own snoring. This paper describes a method to detect oral snoring by extracting the acoustic properties of snoring sounds and using the k-Nearest Neighbor classifier. As a result, over 92% of snoring sounds are successfully classified under the various cross validation evaluations.

Detecting Nonlinear Acoustic Properties of Snoring Sounds using Hilbert-Huang Transform

Since snoring is known to be related to sleep apnea syndrome, many medical/physiological researchers have focused on the biomechanism of snoring and the acoustic properties. Snoring sounds are the mixture of the nonlinear oscillation sounds of the oropharyngeal soft tissues and the airflow noises during inhalation. In conventional studies, however, such properties have not been paid attention to, because there were no suitable methods for the analysis of nonlinear and nonstationary time series data. In this paper, we adopt Hilbert-Huang Transform (HHT) to clarify the nonlinear and nonstationary properties in a nasal snoring sound. As a result, two types of frequency fluctuation are found in the Hilbert-Huang spectrum.

NEURAL CLASSIFICATION OF SNORING SOUNDS FOR THE DETECTION OF ORAL BREATHING DURING SNORING

Healthy people are generally breathing through nose during sleep, but oral breathing will be given rise to if the nasal cavity is gradually being closed. Nasal closure or even nasal congestion leads to open mouth during snoring, which causes the tongue base collapse, the origin of OSAS. Thus, if a simple home device with only a microphone can automatically monitor our snores at bedside and detect oral breathing during snoring, we can perceive an abnormality in our sleep condition easily and early detection and treatment of OSAS will be possible. In our previous work, we proposed some feature extraction methods for the stationary subsequences extracted from oral, nasal, and oronasal snoring sounds and analyzed their acoustic properties in detail. This paper addresses a snoring sound classification based on breathing route during snoring using Multilayer Perceptron, Support Vector Machines, and k-Nearest Neighbor method. According to our experiments, the SVM with Gaussian kernel acquires the best performance where 82.5% of the oral, 89.2% of the nasal, and 73.6 % of the oronasal snoring sounds are successfully classified.

Clustering Ants with Self-Synchronized Interaction

This paper proposes a new interaction mechanism which is similar to selfsynchronization found in Leptothorax species. The task of ants is to gather their food being scattered into one pile without knowing global information. In the alternative models, each agent picks up or drops down his food according to the local environment and does not communicate directly with each other. In our model, nearby agents perform their action in phase by interaction. We demonstrate that such interaction leads to better performance in comparison with non-interaction model.