Considering the Cepstrum of original audio signal as a source of error in the detection process of an echo hiding audio watermarking system, we first analyze the cepstrum coefficients of original signal and adaptively select the coefficients of embedding echoes in each segment. Upon this, we propose three different systems based on single and double echo hiding techniques. We remove the effect of original signal in the detection process, like ISS watermarking, and instead, we consider this effect in the encoding stage. We show that in the absence of channel attacks, the proposed methods have no error in the detection process. We prove that the distortion of the proposed methods is much lower than traditional systems. It is shown that the proposed approaches are robust against common signal processing attacks and have good performance in terms of MOS and SNR values.