Speech Enhancement Models Can Be Backdoored by Their Own Clean Outputs — No Trigger Injection Required

Speech Enhancement Models Can Be Backdoored by Their Own Clean Outputs — No Trigger Injection Required | BedrockNews