Self Voice Conversion as an Attack against Neural Audio Watermarking

TL;DR

This paper explores how self voice conversion can serve as a universal attack that significantly compromises the security and reliability of neural audio watermarking systems by altering acoustic features without changing speaker identity.

Contribution

It introduces self voice conversion as a novel, content-preserving attack method against neural audio watermarking, revealing vulnerabilities in current watermarking techniques.

Findings

Self voice conversion degrades watermark detection accuracy.

The attack preserves speaker identity while altering acoustic features.

Current watermarking methods are vulnerable to this new attack.

Abstract

Audio watermarking embeds auxiliary information into speech while maintaining speaker identity, linguistic content, and perceptual quality. Although recent advances in neural and digital signal processing-based watermarking methods have improved imperceptibility and embedding capacity, robustness is still primarily assessed against conventional distortions such as compression, additive noise, and resampling. However, the rise of deep learning-based attacks introduces novel and significant threats to watermark security. In this work, we investigate self voice conversion as a universal, content-preserving attack against audio watermarking systems. Self voice conversion remaps a speaker's voice to the same identity while altering acoustic characteristics through a voice conversion model. We demonstrate that this attack severely degrades the reliability of state-of-the-art watermarking approaches and highlight its implications for the security of modern audio watermarking techniques.