A Study of Automatic Assessment Methods for Piano Performances

• Main Authors: PHANICHRAKSAPHONG, VARINYA, Varinya Phanichraksaphong
• Other Authors: TSAI, WEI-HO
• Format: Others
• Language: en_US
• Published: 2019
• Online Access: http://ndltd.ncl.edu.tw/handle/6wmmsx

碩士 === 國立臺北科技大學 === 電資國際專班 === 107 === Knowing what the problem is and how to improve the playing skills are crucial for every piano learner. To help piano learners with timely feedback, this research proposes methods to automatic assessment of piano sounds. We analyze piano sounds and classify them into three categories, namely, “Good”, “Normal” and “Bad” in an attempt to assist piano students in knowing what category his/her performance belongs to, thereby knowing how to improve the playing skills with the aid of note and pitch analysis. We define two parts for analysis, one is legato part using vibration note by sustain pedals and the second is staccato part using detached note without sustain pedals. As for the assessment methods, we propose using Gaussian mixture models (GMMs), support vector machine (SVM), and naive Bayes (NB) to classify a piano performance as one of the three categories. Our experiments conducted using 4,680 test samples produced by 13 performers show that the support vector machine is superior to the other methods. Furthermore, to help learners know if some notes of their piano performances are incorrect, we use the Sub-Harmonic Summation (SHS) method to perform pitch estimation, thereby comparing the difference between the pitch produced by a learner and the pitch of ground truth.