In Sensory Transfer project, we develop core technologies for visually impaired people to accurately recognize 3D dynamic surroundings in real time through sensory transfer which deep Learning based high-speed and high-precision Visual Auditorization. This system consists of deep learning-based Semantic-Instance Segmentation, 3D Scene Modeling and Audio AR/MR. We implements the full system in Microsoft Hololens2 and Nvidia Jetson Xavier for portability and efficiency.