The purpose of this study is to use a sound monitoring sensor to judge the activity state of the elderly through the sound. If there is abnormal signal in the voice of the elderly, the family members can be informed to realize abnormal detection in the first time. In this study, we first simulate the old man for help, recorded the experiment of voice and speech respectively, the experimental phonetics and templates than the old man's voice is the voice (reality), if the comparison is same, the alarm will ring, so the template speech database contains a large number of old people in the case of emergency may sound signal. The average success rate of speech matching was 81.67%. This research hopes to help the elderly in their home, so that they can be rescued immediately when they are in emergency and help them out of danger.