In this paper, we present an audio watermarking system based on replica modulation method (RMM). The system uses frequency-shift scheme to create a replica from an original signal. However, differing from previous works we use a part of the input signal to add the replica instead of the whole signal. The most significant part of the signal is extracted first and then RMM is applied to that part. The proposed method makes the system stronger against common audio signal processing such as noise addition or re-sampling. Besides, it has adjustable parameters which allow the system to be adaptable for various data sets. These parameters can be used to control the trade-off between imperceptibility and robustness. This kind of system has a simple design, high quality and good robustness. The more important thing is its strong security inherited from RMM which is immune from synchronization attacks. Also, we will show our implementation and discuss about the experimental results.