In this paper, we propose a real-time perception-level audioto-vibrotactile translation algorithm. Unlike previous signallevel conversion methods, our algorithm considers only perceptual characteristics, such as loudness and roughness, of audio and tactile stimuli. This perception-level approach allows for designing intuitive and explicit conversion models with clear understandings of their perceptual consequences. Our current implementation is tailored to accurate detection of special sound effects to provide well-synchronized audiotactile feedback in immersive applications. We also assessed the performance of our translation algorithm in terms of the detection rate of special sound effects, computational performance, and user preference. All the experimental results supported that our algorithm works well as intended with better performance than the signal-level conversion methods, especially for games. Our algorithm can be easily realized in current products, including mobile devices, gaming devices, and 4D home theater systems, for richer user experience.