Tag
This paper proposes DRL-CLBA, a clean label backdoor attack for speech classification using DDPG reinforcement learning and deep audio steganography, achieving high attack success rates and bypassing several defenses, exposing vulnerabilities in speech-controlled systems.