Top10.1 Introduction
The increase in computational power and the proliferation of the Internet has facilitated the production and distribution of unauthorized copies of multimedia information. As a result, the problem of copyright protection has attracted the interest of the worldwide scientific and the business communities. The most promising solution seems to be the watermarking process where the original data is marked with ownership information hidden in an imperceptible manner in the original signal. Understanding of the human perception processes is the key to successful watermarking. Typical properties of a successful watermarking scheme includes (Cox et al., 2002)
- a.
The watermark should introduce no perceptual distortion.
- b.
The watermark should be embedded into the host signal, rather than into a header.
- c.
The watermark should be hard to remove, or even detect without the prior knowledge of the watermarking scheme and the watermark sequence.
- d.
The watermark should be self-clocking, which also know as synchronization problem.
- e.
The watermark should be readily extracted to completely characterize the copyright owner.
Several techniques in audio watermarking system have been developed in the past decade including lowest-bit coding, phase coding, echo coding, spread spectrum (Bender et al., 1996). Compared to embedding watermark into still images, audio watermarking is much more challenging due to the extreme sensitivity of the human auditory system to changes in the audio signal (Cox et al., 2002). In order to make the embedded watermarks inaudible, a suitable psychoacoustic model is at most of the time, an indispensable part of a good audio watermarking scheme.
Many digital audio watermarking schemes have benefited from the perceptual entropy psychoacoustic model used in several MPEG coding and successfully embedded the watermarks without introducing perceptible distortion (He et al., 2004, 2005; Swanson et al., 1998; S. Jung, et. al., 2003; Garcia, 1999). For example, in (Swanson et al., 1998), the authors first calculate the masking threshold using the psychoacoustic model from MPEG I layer 1 and embed the watermarks to the frequency locations where the signal energy is below the threshold, thus avoiding audible distortion.
S. Jung, et al. (2003) embedded watermarks in the discrete cosine domain (DCT) with the psychoacoustic model from MPEG-2 advanced audio coding (AAC) encoding system. A spectral envelope filter was introduced in the detection phase to reduce the noise variance in the correlation, thus improving the detection bit error rate (BER).
Garcia (1999) proposed a digital audio watermarking scheme using a psychoacoustic model and spread spectrum theory. In this algorithm, the watermarks were locally repeated, interleaved, and spread. A psychoacoustic model was applied to the original audio and the masking thresholds of the audio signal were calculated in the frequency domain. The watermarks were spectrally shaped to fit under the masking threshold of the audio signal.
Seok et al. (2002) proposed an audio watermarking based on traditional direct sequence spread spectrum (DSSS) approach and achieved a watermark insertion rate of 8 bits per second (bps). The inaudibility of the watermark was maintained by incorporating the psychoacoustic model derived from the MPEG I layer I audio coding standard. Their experiments showed the audio watermarking system to be robust to several signal transformations and malicious attacks.