Automatic Detection of Poor Tone Quality in Classical Guitar Playing Using Deep Anomaly Detection Method
Kenta Ogawa, Shun Sawada, Kouichi Katsurada, Hidehumi Ohmura · 2023
Playing the classical guitar requires techniques to control delicate timbres because of its acoustic properties and the characteristics of its nylon strings. Therefore, a consistent fundamental practice focused on single-note playing is important for improving beginners’ classical guitar skills. However, most existing guitar practice support systems are designed for electric and steel-string acoustic guitars and not classical ones. In this study, we propose a system to improve the fundamental skills of classical guitar players. The system automatically detects sounds with poor tone quality caused by an improper playing technique and provides an objective evaluation of single-note performances. In our proposed method, we treat poor-quality sounds that are caused by mistakes while playing as "anomalies" and employing a semi-supervised anomaly detection technique. This approach reduces the effort required to collect and label poor tone quality data and enables the detection of various types of degraded tone quality. Verification experiments confirmed the high accuracy of the proposed model in detecting poor tone quality in classical guitar playing, demonstrating its potential as an effective practice support for classical guitarists.