Article: Neural ordinary differential equation waveform modelling for multidimensional analysis of singing technique Journal: International Journal of Information and Communication Technology (IJICT) 2026 Vol.27 No.93 pp.21 - 45 Abstract: The precise analysis of singing techniques is of vital importance for intelligent vocal education. The existing methods process singing by segmentation, which destroys the characteristics of continuous changes of techniques, resulting in poor recognition of dynamic techniques such as vibrato. To address this issue, this paper proposes a neural ordinary differential equations waveform modelling network. This framework treats singing as a continuous process and simulates the physical laws of vocal cord vibration, enabling the model to more naturally capture the subtle changes in techniques. Experiments show that the comprehensive recognition accuracy of this method reaches 0.874, and the comprehensive evaluation index reaches 0.932. Compared with the current optimal method, it has improved by 4.2% and 3.1% respectively. The recognition of trills has also improved by 12.3%. This research provides high-precision and interpretable technical support for intelligent vocal education. Inderscience Publishers - linking academia, business and industry through research

Title: Neural ordinary differential equation waveform modelling for multidimensional analysis of singing technique

Authors: Yilin Zhao

Addresses: College of Aesthetic Education, Henan Vocational College of Light Industry, Zhengzhou, 450000, China

Abstract: The precise analysis of singing techniques is of vital importance for intelligent vocal education. The existing methods process singing by segmentation, which destroys the characteristics of continuous changes of techniques, resulting in poor recognition of dynamic techniques such as vibrato. To address this issue, this paper proposes a neural ordinary differential equations waveform modelling network. This framework treats singing as a continuous process and simulates the physical laws of vocal cord vibration, enabling the model to more naturally capture the subtle changes in techniques. Experiments show that the comprehensive recognition accuracy of this method reaches 0.874, and the comprehensive evaluation index reaches 0.932. Compared with the current optimal method, it has improved by 4.2% and 3.1% respectively. The recognition of trills has also improved by 12.3%. This research provides high-precision and interpretable technical support for intelligent vocal education.

Keywords: neural ordinary differential equations; singing technique; physics-informed machine learning; continuous-time dynamic modelling.

DOI: 10.1504/IJICT.2026.155874

International Journal of Information and Communication Technology, 2026 Vol.27 No.93, pp.21 - 45

Received: 03 Apr 2026
Accepted: 08 May 2026
Published online: 24 Aug 2026 *

Title: Neural ordinary differential equation waveform modelling for multidimensional analysis of singing technique

Keep up-to-date