In this paper, we propose a modified version of echo hiding in sub-band signals, which employs multiple echo kernel. This method allows reducing the distortion of the original audio signal, with respect to both subjective distortion (perceptibility) and objective distortion such as SNR(Signal to Noise Ratio). Through the experimental results, we proved that the use of multiple echo kernel gives good performances, in terms of SNR and detection rate of watermark, compared to the one of single echo kernel.