In this paper, a novel of high imperceptibility and capacity blind audio watermarking system based on Rounding Reduced-Arc M-Ary Phase Shift Keying (MPSK) which is optimized by the Genetic Algorithm is explained. The audio file¡¯s sample will be divided into several frames and then transformed into frequency domain using FFT. The watermark data will be embedded into the phase element of the selected samples that have been framed and transformed. We use two modulation schemes for the embedding process, 256-PSK and BPSK to compare the effects of the number of M-ary used in the system. To improve the system, we used the Genetic Algorithm to find the best embedding parameter to yields the most optimal output based on the Perceptual quality of watermarked audio and the robustness of the watermark. Experimental results show that the proposed audio watermarking system produces high quality of watermarked audio while loading huge amount of watermark data.