情報処理学会 インタラクション2021

The increasing importance of sleeping quality and the awareness of sleep related disorders have led to emerging research in the field of wearable sensor with the purpose to make the sleep sensing become more comfortable and accessible. Among various methods, the use of audio sensor is known to be one of the most direct approach. However, further research using audio sensor to detect sleeping-state and apnea should be done to explore various new contexts and possibilities. Thus, this study proposes a wearable system to recognize human contexts of breathing, swallowing, body movement, and oral sound for the further use of sleeping-state and apnea severity detection. Several audio data combination methods of Aggregation Methods and Stacking Methods were proposed to improve the accuracy of the context detection. The Stacking Method with Support Vector Machine Polynomial Kernel as both first and second level classification resulted the best performance of 85.1\% accuracy and 18\% average improvement.