Watermarking of speech signal through phase quantization of sinusoidal model

In this paper, we present a blind speech watermarking scheme that uses the sinusoidal model of speech signal for watermarking. The basic idea is that the human auditory system is insensitive to absolute phase, making it a good candidate for data hiding. In the proposed approach, speech data is initially partitioned into overlapped frames. Sinusoidal analysis was performed on each frame and the watermark data embedded in the phase section of the appropriate component sine wave. Experimental results show that the proposed watermarking algorithm is inaudible and robust against common attacks such as adding white Gaussian noise, filtering, re-sampling, re-quantization etc.