Abstract:  Aiming at the problem of sudden changes or inconsistent loudness at the receiving end caused by factors such as interference and unstable audio sources in the voice communication system.A new VSSLMS-AGC (variable step size least mean square-adaptive gain control) method is proposed. First the method uses exponentially weighted moving average to record historical gradient information.This can adjust the step size of the least mean square algorithm, to make the algorithm converge faster and stabilize the error smaller.Then the variable step size idea is applied to automatic gain control,to ensure that the voice gain value can be adjusted adaptively.so that the output range of the original audio signal with different amplitudes is constant without losing the substantial characteristics,The method can reduce the sound drop and improve the auditory experience.