The authors propose a table-top with virtual keyboards for multi-users to work in a collaborative environment. The proposed system has two main modules: a system for virtual keyboards with touch event detection from depth data of a Kinect and a word predicting module based on the idea of Hidden Markov Model and Trie data structure. The system can replace physical keyboards, improve the accuracy of a virtual keyboard, and increase the typing speed of users. Our experimental results show that our system archives an accuracy of 94.416% with the virtual keyboard, saves 11-22% of keystrokes, and corrects 89.02% of typing mistakes.