Yuta Nishikawa

Numerous systems have been developed for the automatic evaluation and feedback of electric guitar performance. However, these systems have a significant limitation: they consider only "timing" and "pitch" in their evaluation. In practice, human evaluation involves a much broader range of factors. To address this issue, previous studies have proposed automatic evaluation methods that incorporate additional criteria. Nevertheless, these methods remain unable to evaluate sounds produced using specialized electric guitar techniques. In this study, we propose a method for automatic proficiency estimation of vibrato on electric guitar. Specifically, we extracted acoustic features focusing on peaks of the Mel fundamental frequency, including the number of peaks, mean and variance of peak width, and mean and variance of peak height. We then regressed the evaluation scores against these extracted acoustic features using a Relevance Vector Machine (RVM). As a result, we achieved a coefficient of determination of 0.723. This result indicates that the extracted features are highly correlated with human evaluation scores and enables a tentative automatic evaluation of vibrato sound on electric guitar.