This paper proposed automated robot speech gesture generation system for service/entertainment robots. This system can automatically generate robot Beat gestures accompanied by speech interaction situation with humans. Beat gestures have no specific semantic meanings to communicate, but it is commonly believed that they are an essential factor of natural communication. We extracted basic gesture patterns by analyzing human speech videos and built a correlation model between gesture pattern and punctuation mark of speech sentence. This model can select a gesture pattern sequence to make a robot gesture expression motion for arbitrary input speech sentence, and synchronize it with vocal wave file from TTS(Text-To-Speech) engine.