To improve drum performance, it is important to consider the volume balance of the bass drum, snare drum, and hi-hat. However, it is difficult for players to evaluate the balance of each volume of these instruments while playing. In addition, while it is possible to self-diagnose by recording drum performances, it is not always efficient to re-record and re-play drum performances based on the correction points. Therefore, we developed a system that uses semi-supervised non-negative matrix factorization (SSNMF) to separate a player’s drum performance, recorded with a unidirectional microphone installed in front of the drum kit, into the bass drum, snare drum, and hi-hat sound sources in real-time, and estimates each volume at the time of beating. In addition, the system also visualizes their volume balance and enables the player to control the power of beating. We experimented using this system for actual drum performance and clarified its usefulness and points for improvement based on the feedback obtained from the experiment participants.