Data Hiding for Stereo Audio Signals

Introduction

Recent advances in high-speed digital communication networks, most notably the Internet, have caused significant changes in how we address digital media. It has become quite practical to use these networks to access digital media on demand. We can also easily store, copy, edit, and distribute these media. Because digital processing of digital content is possible with virtually no degradation, flawless copies of copyrighted content can be distributed to a large number of users instantly through these networks. This situation has called for the protection of intellectual ownership and prevention of unauthorized tampering of digital media content. One solution is to digitally hide copyright information within digital content without altering its quality, i.e., digital contents are “marked” with copyright information transparently.

Historically, the volume of research on data hiding (commonly known as “watermarking,” which refers to copyright data hiding in host signals) has been for images and videos (Bender, Gruhl, Morimoto, & Lu, 1996). However, we have also recently developed new algorithms for information hiding in speech and audio (Cvejic & Seppänen, 2008). Most of these new algorithms take advantage of the Human Auditory System (HAS) to hide information into host speech or audio signals without causing significant perceptual disturbances. Two properties of HAS that are frequently utilized in information hiding for audio are temporal masking (Elliot, 1969) and spectral masking (Zwicker, 1982). These properties are also used in many MPEG audio coding standards (Gibson, Berger, Lookabaugh, Lindberg, & Baker, 1998; ISO/IEC JTC1/SC29/WG11 11172-3, 1999; ISO/IEC JTC1/SC29 13818-7, 2006).

Although there are numerous audio data-hiding algorithms, few are geared toward stereo host audio signals. The redundancy in stereo channels may be exploited to hide data transparently. In this chapter, we will show that it is indeed possible to exploit this redundancy and embed data without significant degradation of the host signal.

In the first part of this chapter, we propose a stereo audio data-hiding algorithm that hides data in a host signal by introducing data-dependent delays to the host signal’s high-frequency stereo channels. Because the HAS is known to be relatively insensitive to fine structures in the high-frequency region, the high-frequency portion is replaced with a single mid-channel. The relative delay between the left and right channel is controlled by the hidden data. Although blind detection (detection without the original signal) is shown to be possible, and robustness to common disturbances is experimentally shown, the audio quality degradation of the embedded audio is shown to be significant for many of the sources tested.

In the latter part of this chapter, we propose a data-hiding algorithm that uses the property of the HAS mentioned above to code embedded data in the relative polarity of the added echo between two stereo channels. We add this echo to control the correlation between channels and match this correlation to the original signal, making the alteration more difficult to perceive. This interchannel correlation is known to influence the stereo image “width” of stationary auditory objects. We show that blind detection is also possible. We also show that this algorithm is robust to most of the disturbances tested, and the data-embedded host audio is “good” to “excellent” in terms of quality.