It has been emphasized to comprehensively improve the four skills of English; reading, writing, listening, and speaking in the university English education in Japan. However, the teaching of speaking skill has not been effectively carried out because of the insufficient ability of teachers and ineffective teaching materials. It has been expected to acquire practical English proficiency these days and so effective teaching materials are required for teaching the ability to speak. In this research, I developed the Shikibu as an e-learning system for English speaking practice by combining speech recognition and artificial intelligence. I introduce the overview of the development of this English conversation practice system aiming at improving speaking ability using response and utterance practice. I will also summarize the responses and impressions of the system from the users and report its effectiveness.