A real-time unusual voice detector based on nursing at home

In this paper, we will propose a method to detect an unusual voice for nursing system. Based on the healthy condition of a person, we define four kinds of unusual voices including cough, groan, wheeze and cry for help. When the person nursed sends out the unusual voices, we judge that his health condition have a doubt, and need someone to pay attention. In order to detect the unusual voices, we extract five features on audio waveform, including the number of segmented parts, duration of waveform, mean of volume, zero crossing rate and correlation. Experimental results show that the detection rate is 94%~97% for these four kinds of unusual voices. In false alarm, there are only 0.08% of wrong rates.