With the development of speech synthesis and lip synch, it is easy to impersonate and create fraud audio and video contents by machine learning method from voice samples. The machenism can be applied to video data, since video data including both of audio and image data in frames. To prevent the voice print from being stolen and abused, adding noise in sound has been proposed in the conventional works, however, distortion is a challenge. In this paper, a novel method for noise perturbation in specific frequency domain is proposed to achieve both of the imperceptibility and confidentiality. Laplace noise with particular security utility are generated and perturbed to specific frequency domain of speech data. The proposed method has appropriate effectiveness to degrade the speaker verification by reproducing the voice print theoretically, and the audio quality of data generated by the proposed watermarking method is imperceptible with an average segSNR 22.23 dB, and MOSLQO 4.27.