Synchronization Insensitive Audio Watermarking: Robust Against Random Cropping

Recent years have seen increased Internet traffic and the proliferation of digital audio distribution in the form of MP3 files. As such, copyright protection of digital audio works is becoming increasingly important. As a complement to conventional encryption techniques, watermarking provides powerful tools for copyright protection and has become a very active research area in recent years.

Audio watermarking is a technique that embeds information with specific meaning into the host media without interference to the quality of the original work. Proposed applications of audio watermarking include copyright protection, annotation, authentication, broadcast monitoring, and tamper proofing. Depending on the application, the inserted watermark data can include copyright information, serial numbers, text (e.g., the name of the composer and title of the work), a small image, or even a small clip of audio. The watermark is hidden in host media and is usually imperceptible to humans. The watermark must also be able to withstand signal-processing manipulations.

In general, a good audio watermarking scheme should possess the following properties:

• • The watermark must be embedded into the host media rather than stored in a file header or a separated file, or else it can be removed or modified easily.

• • The watermark should not introduce any perceptible artifacts that affect the quality of the original signal (i.e., perceptual transparency).

• • The watermark should not be detected without prior knowledge of the watermark sequence. (To ensure the security, a secret key must be used for the embedding and detection process.)

• • The watermark should be robust against common signal processing techniques such as lossy compression, filtering, resampling, noise addition, etc.

• • The computational cost of embedding and detection should be low enough for real-time processing.

• • The watermark should be self-clocking to ensure its recovery when facing malicious cropping or time-scale modification.

• • Under most circumstances, the watermark should be detected without resorting to the original audio, which is often very difficult to find in an open environment like the Internet.

• • Finally, the watermarking algorithm should be public; that is, the security depends on the secret key but not the secrecy of the algorithm.

It is very difficult to design a watermarking system that meets all these requirements. Some properties such as robustness, transparency, and data capacity conflict with each other. The goal is achieving the best trade-off among them.

In this article, we propose a novel audio watermarking scheme based on statistical features in the wavelet domain. In each audio frame, the mean value of the wavelet coefficients at the coarsest approximation subband is calculated as the statistical feature, which is supposed to be invariant to a wide range of attacks. The basic idea is to embed watermark data by transforming this feature to a given positive or negative number, one bit per frame. The experimental results demonstrate that this algorithm is robust to common audio signal processing such as MP3 compression, low-pass filtering, equalization, echo addition, resampling, and noise addition. Furthermore, it also shows certain robustness to synchronization attacks like random cropping and time-scale modifications. The watermarked audio has very high perceptual quality and is indistinguishable from the original signal. A blind watermark detection technique without resorting to the original signal is developed to identify the embedded watermark under various types of attacks. To ensure the security of the watermark, a random chaotic sequence is employed in the process of embedding and detection.

In this research, special attention is paid to the synchronization attack caused by casual audio editing or malicious random cropping, which is a low-cost yet effective attack to most existing watermarking algorithms based on the classical spread-spectrum technique. This new approach is more robust when the original signal is not available, and it is very insensitive to the change of synchronization structure.

The concept of random cropping in audio watermarking is quite different from that used in most of the image-watermarking literature, where the cropped image is usually restored to the original size of standard test images before watermark detection. With audio watermarking, however, it is very difficult to know the exact length of the...