DARE: Deceiving Audio–Visual speech Recognition model

作者：

Highlights：

• We initiate a targeted attack on AVSR model and detection network simultaneously.

• Our attack shows promising results on the publicly available well known LRW dataset.

• We successfully circumvent popular defences while maintaining imperceptibility.

摘要

•We initiate a targeted attack on AVSR model and detection network simultaneously.•Our attack shows promising results on the publicly available well known LRW dataset.•We successfully circumvent popular defences while maintaining imperceptibility.

论文关键词：Audio–Visual Speech Recognition,Adversarial attacks,Cross-modality,Detection network

论文评审过程：Received 13 April 2021, Revised 19 August 2021, Accepted 15 September 2021, Available online 20 September 2021, Version of Record 27 September 2021.

论文官网地址：https://doi.org/10.1016/j.knosys.2021.107503