An audio mixing method, apparatus and system, which can ensure sound quality after audio mixing and reduce consumption of computing resources. The method includes: receiving an audio stream of each site, and analyzing the audio stream of each site to obtain a sound characteristic value of a sound source object; selecting, according to a descending sequence of sound characteristic values of sound source objects, a predetermined number of sound source objects from the sound source objects to serve as main sound source objects; determining, according to a relationship between a target site and the sites where the main sound source objects are located, audio streams that require audio mixing for the target site; and performing audio mixing on the audio streams that require audio mixing for the target site and sending the audio streams after the audio mixing to the target site.