XAttnMark: Learning Robust Audio Watermarking with Cross-Attention
Researchers have developed XAttnMark, a novel audio watermarking system designed to combat copyright infringement and misinformation from generative AI. This system utilizes a cross-attention mechanism and temporal conditioning to improve both the robustness of watermark detection and the accuracy of attribution. XAttnMark also incorporates a psychoacoustic-aligned loss function to enhance watermark imperceptibility, demonstrating state-of-the-art performance against various audio transformations and generative editing. AI
IMPACT Enhances methods for detecting and attributing AI-generated audio, aiding in intellectual property protection and combating misinformation.