Deployment of an AI application software in piano practice: Unheard and unaware musical mistakes

In musical practice, realizing one’s mistakes or listening to all the sounds produced properly is not always possible. We developed an AI application software that can detect mistakes in a recorded performance and show the user how to practice and rectify these errors. This study aimed to investigate whether AI applications help learners practice piano and improve their performance. Sixteen undergraduate and postgraduate students (two males, mean age = 20.38 years) majoring in music at a university in Tokyo participated in this study. After a trial of the application, the participants were asked to practice W. F. Bach’s Allegro in A Major for 5 min and then record their performance. This was repeated five times. Performance data for all five sessions were recorded using a camera and exported as MIDI audio data. A total of 2745 errors were detected in the performance data and classified into 11 categories: pitch errors, omissions (whole), omissions (part), added ties, timing errors, score-based insertions, empty insertions, beat insertions, repetitions, returns, and replays. “Pitch errors” were the most frequent (605 errors), followed by “score-based insertions” (519 errors). In the questionnaire, the participants reported that they believed the AI application would be useful in identifying mistakes that they would not have been able to detect themselves and in improving their performance with further practice.