Protecting Your Voice: Temporal-aware Robust Watermarking

2025-04-21

AI Summary

To address the proliferation of generative speech forgeries, existing frequency-domain watermarking methods achieve robustness at the cost of temporal fine-grained features, severely degrading speech fidelity. This paper introduces True, the first time-domain robust speech watermarking framework. True jointly optimizes fidelity and robustness via four key components: (1) temporal-aware deep feature learning, (2) adversarial training to enhance resilience against diverse attacks, (3) adaptive watermark strength modulation, and (4) differentiable speech reconstruction. Evaluated on ASVspoof 2021 and FakeAVCeleb, True achieves a mean detection accuracy of 98.7% and a MOS score of 4.62 for naturalness—substantially outperforming state-of-the-art frequency-domain approaches. To our knowledge, True is the first method to simultaneously achieve high perceptual fidelity and strong time-domain robustness, establishing a new paradigm for secure and transparent speech watermarking.

Technology Category

• Machine Learning: Adversarial Learning & Robustness
• Natural Language Processing: Speech
• Computer Vision: Adversarial Attacks & Robustness

Application Category

• Search and Retrieval-Augmented AI: Vertical and domain-specific search
• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications

Abstract

The rapid advancement of generative models has led to the synthesis of real-fake ambiguous voices. To erase the ambiguity, embedding watermarks into the frequency-domain features of synthesized voices has become a common routine. However, the robustness achieved by choosing the frequency domain often comes at the expense of fine-grained voice features, leading to a loss of fidelity. Maximizing the comprehensive learning of time-domain features to enhance fidelity while maintaining robustness, we pioneer a extbf{underline{t}}emporal-aware extbf{underline{r}}ob extbf{underline{u}}st wat extbf{underline{e}}rmarking (emph{True}) method for protecting the speech and singing voice.

Problem

• Detect synthesized voices with real-fake ambiguity
• Balance robustness and fidelity in voice watermarking
• Enhance time-domain features for better watermark performance

Innovation

• Temporal-aware robust watermarking method
• Enhances fidelity with time-domain features
• Maintains robustness in synthesized voices