This paper presents a real-time accompaniment system for the sung voice. A sung voice recognition algorithm has been developed to extract the score info concerning with a performance from the inputted signals captured by a microphone. The algorithm can effectively detect pitches of the sung voice and give a proper score segmentation result from the mixture of both sung voice and accompaniment by analyzing the spectrum peaks of the inputted signals. The experimental results show that the accompaniment system developed can produce a real-time accompaniment with good effect for the sung voice.