MARC: Multi-Bit Watermarking for Autoregressive Audio Generation against Codec Attacks
MARC embeds multi-bit watermarks in autoregressive audio that survive codec and overwriting attacks.
Researchers propose MARC, a multi-bit generative watermark for autoregressive audio generation. It joins intrinsic token representations with confusion patterns from retokenization and multiple codecs, then uses payload-driven cluster scheduling to embed bits. On speech, dialogue, and music, average bit-extraction accuracy is 97.3% for unmodified watermarked audio, and the watermark remains extractable under codec and overwriting attacks.
- Builds a codec-aware token-cluster space from retokenization and multiple codecs
- Embeds multi-bit payloads instead of zero-bit presence checks
- 97.3% average bit-extraction accuracy on unmodified watermarked audio
- Extraction still works under diverse codec and overwriting attacks
From the original article
1 min read at arxiv.orgGenerated audio is now used in a range of applications, creating a need to verify its origin after distribution and signal processing. This task is particularly challenging for autoregressive audio generation because codec processing can alter the token sequence recovered from the waveform. Such changes reduce the reliability of watermark detection and payload decoding.
By Liaoran Xu · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.