Abstract: It is difficult for artificial cochlear implant（CI） children to master and use traditional speech rehabilitation training systems based on WIMP interface paradigm. In order to solve this problem, a tangible user interface-based speech rehabilitation training system is proposed in this paper. The Microsoft Kinect is utilized to track the puppets which are familiar to CI children to interact with computers, and the ICP（iterative closest point） algorithm is optimized for the real-time tracking of the relative small puppets. Through the proposed method, speech rehabilitation training tasks are designed and developed corresponding to the typical speech rehabilitation training process. The user studies of CI-children usage and questionnaire investigations of instructors have been analyzed. Experimental results prove the feasibility and effectiveness of the proposed system.