Hierarchical Approach to Detect Common Mistakes of Beginner Flute Players

Music lessons are a repetitive process of giving feedback on a student’s performance techniques. The manner in which performance skills are improved depends on the particular instrument, and therefore, it is important to consider the unique characteristics of the target instrument. In this paper, we investigate the common mistakes of beginner flute players and propose a hierarchical approach to detect such mistakes. We first examine the structure and mechanism of the flute, and define several types of common mistakes that can be caused by incorrect assembly, poor blowing skills, or mis-fingering. We propose tailored algorithms for detecting each case by combining deterministic signal processing and deep learning, to quantify the quality of a flute sound. The system is structured hierarchically, as mis-fingering detection requires the input sound to be correctly assembled and blown to discriminate minor sound difference. Experimental results show that it is possible to identify different mistakes in flute performance using our proposed algorithms.