Silence is Golden: Leveraging Adversarial Examples to Nullify Audio Control in LDM-based Talking-Head Generation

Yuan Gan, Jiaxu Miao, Yunze Wang, Yi Yang · 2025

Read Silence is Golden: Leveraging Adversarial Examples to Nullify Audio Control in LDM-based Talking-Head Generation by Yuan Gan, Jiaxu Miao, Yunze Wang, Yi Yang (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2506-01591.txt.

Description

Yuan Gan, Jiaxu Miao, Yunze Wang, and Yi Yang build a shield for portrait photos against a new kind of fake video, in a 2025 paper. Modern tools can animate a still face so it mouths along to any audio track, well enough to fool viewers. Such fakes feed scams and false news. Earlier shields added noise patterns to portraits, but they could not stop the audio from driving the face, and cleaning tools could wash the noise away.

Their method, Silencer, works in two stages. First, a training goal teaches the pattern to make the generator ignore the audio track, leaving the face still. Second, a further goal hardens the pattern against the cleaning steps an attacker would use. Tests show the shield holds where earlier ones failed. The authors frame the work as a privacy defense and a warning about talking-head tools.

Catalog record

CREATORYuan Gan, Jiaxu Miao, Yunze Wang, Yi Yang
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE48 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2506-01591

Browse the full library · About the Tower of Babel