Transfer Learning in Vocal Education: Technical Evaluation of Limited Samples Describing Mezzo-soprano

Abstract Vocal education in the music field is challenging to quantify due to the individual variability in singers' voices and the diverse criteria for evaluating singing techniques. Deep learning offers substantial potential in music education by efficiently handling complex data and performing quantitative analyses. However, achieving accurate evaluations with limited samples, especially for rare vocal types such as Mezzo-soprano, requires extensive well-annotated data to support deep learning models. To address this challenge, we employ transfer learning by leveraging deep learning models pre-trained on the ImageNet and Urbansound8k datasets, aiming to enhance the precision of vocal technique evaluation. Additionally, we address the issue of data scarcity by constructing a dedicated dataset, the Mezzo-soprano Vocal Set (MVS), specifically designed for vocal technique assessment. Our experimental results demonstrate that transfer learning improves the overall accuracy (OAcc) of all models by an average of 8.3%, with the highest accuracy reaching 94.2%. Beyond technical improvements, this study introduces a new quantitative assessment method for music education. By bridging deep learning methods with vocal pedagogy, our approach offers significant practical implications for enabling more objective and personalized vocal education, and advancing the scientific understanding of singing techniques.