Through exploiting the human perception of spatial sound, a new approach for compression coding of multi-channel audio signal based on ITU-T G.719 codec is put forward in this paper. Multi-channel input signals are converted to a down-mixed signal plus spatial perceptual parameters by use of down-mix and up-mix step-by-step techniques in frequency domain. The algorithm can significantly reduce the coding rate under the premise of an acceptable sound quality in combination with the G.719 audio codec. The paper presents the implementation of the algorithm and describes in detail the calculation and features of the selected spatial parameters. Finally some experiments are done to evaluate the algorithm from the perspective of the compression ratio, the reconstructed sound quality, and the algorithm complexity.