StirMark Benchmark: Audio watermarking attacks based on lossy compression

StirMark Benchmark is a well-known evaluation tool for watermarking robustness. Additional attacks are added to it continuously. To enable application based evaluation, in our paper we address attacks against audio watermarks based on lossy audio compression algorithms to be included in the test environment. We discuss the effect of different lossy compression algorithms like MPEG-2 audio Layer 3, Ogg or VQF on a selection of audio test data. Our focus is on changes regarding the basic characteristics of the audio data like spectrum or average power and on removal of embedded watermarks. Furthermore we compare results of different watermarking algorithms and show that lossy compression is still a challenge for most of them. There are two strategies for adding evaluation of robustness against lossy compression to StirMark Benchmark: (a) use of existing free compression algorithms (b) implementation of a generic lossy compression simulation. We discuss how such a model can be implemented based on the results of our tests. This method is less complex, as no real psycho acoustic model has to be applied. Our model can be used for audio watermarking evaluation of numerous application fields. As an example, we describe its importance for e-commerce applications with watermarking security.