Regarding the automatic transcription, studies on multi-pitch estimation have been pursued for a long time. However, most studies have not implemented a system that can endure practical use at present. It is not solved by the performance gain of the computer because its difficulty is not only the problem of computational complexity. Therefore we propose a transcription-aid system that cooperates with a user without entrusting all. In this paper, the user inputs the highest note in the harmony, and the system cuts the bandwidth over its fundamental frequency. Moreover, higher accuracy is achieved by using the music knowledge of the chord progression. We experimented on estimating the fundamental frequencies of MIDI and actual audio signals and confirmed the effectiveness of these methods.
