AudioAuth: A Dual-Watermarking Framework for Robust Audio Integrity and Source Attribution
Summary
No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.
Summarise with VIRAT Free to try with your Science Catalogs account.Abstract
Voice biometric systems face escalating threats from neural speech synthesis technologies capable of producing speaker-indistinguishable audio. Existing audio watermarking defenses remain vulnerable to spectral-domain attacks, offer only coarse tamper localization, and often degrade under codec compression. We propose AudioAuth, a dual-watermarking framework for joint content integrity verification and source attribution in voice biometric security, extending our prior single-channel work WaveVerify. AudioAuth employs a frequency-partitioned dual-channel encoding strategy: a fixed model identifier embedded in even frequency bands enables reliable source attribution, while a dynamic payload embedded in odd bands supports fine-grained temporal tamper localization. A dynamic effect scheduler further enables co-adaptive training, improving robustness against adversarial distortions and signal transformations. Evaluated against six state-of-the-art baselines spanning signal-level (AudioSeal, WavMark, WaveVerify) and representation-level (WMCodec, Timbre, DiscreteWM) approaches, AudioAuth achieves precise temporal localization (MIoU = 0.983), accurate multi-source TTS attribution (98.8% across eight systems), and negligible impact on speaker verification performance (EER increase of only 0.07%).
Fields of Study & Subjects
Advanced Steganography and Watermarking Techniques
Speech Recognition and Synthesis
Music and Audio Processing
Publication details
- Journal / source
- IEEE Transactions on Biometrics Behavior and Identity Science
- Volume
- 8
- Issue
- 5
- Pages
- 680–694
- Publication date
- 30 March 2026
- Language
- EN