When the spatial orientation of audio object changes rapidly, the human brain will pay attention to audio object. Traditional audio attention model cannot effectively express the situation when the spatial orientation of audio object changes rapidly. In view of the limitations of the existing audio attention model, the self-information of the variation in spatial parameter based audio attention model is proposed. The variation of interaural level difference is used to characterize the variation in spatial orientation of audio object, and the probability matrix of the interaural level difference of each sub-band in audio frame is obtained by off-line statistics of substantial audio training sequences. From the probability matrix and the calculation formula of self-information, the self-information of the variation in interaural level difference of each sub-band in the current frame is obtained. Finally, the audio attention model is established by summing the self-information. After theoretical analysis of the proposed audio attention model, the feasibility of the model is proved.