Propose a new strategy to cope with impulsive noise in audio signal based on time-frequecy analysis. This method detects impulsive noise by distinguishing the difference of frequency characteristics and correlations between the clean and noisy signal based on the spectrogram of the corrupted audio. Introduces a scheme, “band-pass”, for the first time, and use median filter with a filtering system to suppress and remove short time and long time impulsive noises. For the sake of evaluation of a denoising strategy, further propose four criterions specially to assess the performance of various strategies compared with SNR. Based on the four criterions, firstly analyze how to adjust parameters to achieve the best performance，then extensive experiments are performed to confirm the feasibility and achievable performance of the proposed method. Finally give some insights of this novel strategy according to the simulation results.