In this paper, we propose a modular role player game-learning system based on voice recognition for language education in cloud. Each designed modular language teaching materials can be embedded into the game platform on the virtual machine in cloud. By connecting to the external voice recognition resource, the pronounce of learners can be recognized. Only when the voice recognition result data (in words) of learners matches the on demand answer, the learners can be identified and enter the further learning unit. By designing different scenarios, different vocabularies or conversations can be added into the RPG game. The information about users in the individual game learning system can be stored in the cloud database. In addition, the whole learning system is established as a virtual machine. System maintainer can configure the learning system easily and quickly. Based on cloud, different remote devices can connect to server for learning.