The transmission rate of the Internet is increasing. As a result, the real-time streaming system has become possible over the Internet. Current Internet applications need error-free transmission. However, the real-time streaming system has a limitation at the transmission completion time because the system is a communication tool. To formulate the quality at the portable telephone level, one way transmission delay is assumed to be 150ms so the real-time streaming system needs transmission with fewer error at a time. Meanwhile, various communication protocols are developed. However, the performance of protocols is influenced in the network state. Therefore, a suitable protocol is different by the network state. In this paper, we propose the real-time streaming system that selects a suitable protocol by using reinforcement learning in the network state.