Earthquake is a kind of natural disaster with strong destructive power. As a kind of earthquake precursor signal, earthquake sound is of great significance of earthquake prediction. According to the characteristics of the earthquake sound signal, the earthquake sound monitoring data ten days before the earthquake was analysed using wavelet technology, the earthquake sound monitoring signals of different frequency bands was compared, and the change rule of the earthquake sound signal within ten days before the earthquake was emphatically studied. It was concluded that: the infrasound wave signal with normal fluctuation changed regularly into time; Ten days before the earthquake, the peak mutation of infrasonic signal occurred, which increased exponentially compared with the mean peak mutation value of earthquake sound signal. The number of this abnormal phenomenon increased to the shortening of the impending earthquake time, especially five days before the earthquake; The intensity of the earthquake sound signal in the frequency band of the infrasonic wave is gradually decreasing from the increase in the signal frequency. By studying the variation characteristics of earthquake sound signal with time before earthquake, it is of certain reference significance to study the correlation between abnormal ground earthquake signal and earthquake in earthquake prediction.