Reshaping Representation Space to Balance the Safety and Over-rejection in Large Audio Language Models

Open in new window