DeepMark Benchmark: Redefining Audio Watermarking Robustness

This paper introduces DeepMark Benchmark, a comprehensive and extensible framework for evaluating the robustness of audio watermarking algorithms. The benchmark enables systematic evaluation of watermarking methods against a diverse range of attacks, including audio editing operations, distortion and desynchronization attacks, end-to-end transmission scenarios, and deep learning-based transformations leveraging generative models and neural audio processing. Using this framework, we benchmark several state-of-the-art audio watermarking models and provide a comparative analysis of their robustness across attack categories. In addition, we introduce Process Disruption Attacks, which occur when multiple watermarking models are applied to the same audio signal, leading to interference between watermarking mechanisms without requiring prior knowledge of the system architecture. The framework is designed to be modular and easily extensible, enabling future integration of new watermarking methods and attack strategies. It is intended to support both academic research and industrial validation of robust audio provenance solutions. Code is available at: https://github.com/deepmark/deepmarkpy-benchmark.