The generation of musical scores directly from a traditional Japanese stringed instrument (it is called “Koto” in Japanese) performance recordings, without the need for manual modification, not only facilitates the learning process for Koto players but also contributes to the preservation and advancement of Japanese musical instrument culture. While automatic music transcription has been studied extensively for various instruments, there has been little to no research focusing specifically on Japanese traditional instruments. We approach automatic music scoring as a two-stage process: melody recognition and score generation. Melody recognition involves extracting a sequence of pitches, note durations, and onset times. This paper proposes a melody recognition method for solo Koto performances with limited simple playing techniques, aiming to achieve high accuracy by integrating signal processing and machine learning.