Article: Improving the accuracy of English pronunciation using graph neural networks and transfer learning Journal: International Journal of Information and Communication Technology (IJICT) 2026 Vol.27 No.83 pp.41 - 59 Abstract: The automatic correction of English pronunciation has long been hindered by two major problems: extremely scarce non-native annotated data, and the difficulty for traditional models to effectively capture the pronunciation similarities between phonemes. To address these issues, this paper proposes a pronunciation correction framework named pronunciation correction with graph and transfer, which structures the phonemes into a relational graph to utilise its dependency information, and leverages cross-language transfer learning to reduce reliance on target annotated data. On public datasets, this method achieves area under the curve scores of 0.862 and 0.879 respectively, which are 7.3% and 6.8% higher than the baseline model; in precision metrics, it achieves improvements of 8.1% and 7.4%. The results show that pronunciation correction with graph and transfer significantly enhances the accuracy of pronunciation error detection, providing a feasible path for computer-assisted pronunciation training in data-scarce scenarios. Inderscience Publishers - linking academia, business and industry through research

Title: Improving the accuracy of English pronunciation using graph neural networks and transfer learning

Authors: Yilin Zhang

Addresses: School of Business Administration, Qingdao Vocational and Technical College of Hotel Management, Qingdao, 266100, China

Abstract: The automatic correction of English pronunciation has long been hindered by two major problems: extremely scarce non-native annotated data, and the difficulty for traditional models to effectively capture the pronunciation similarities between phonemes. To address these issues, this paper proposes a pronunciation correction framework named pronunciation correction with graph and transfer, which structures the phonemes into a relational graph to utilise its dependency information, and leverages cross-language transfer learning to reduce reliance on target annotated data. On public datasets, this method achieves area under the curve scores of 0.862 and 0.879 respectively, which are 7.3% and 6.8% higher than the baseline model; in precision metrics, it achieves improvements of 8.1% and 7.4%. The results show that pronunciation correction with graph and transfer significantly enhances the accuracy of pronunciation error detection, providing a feasible path for computer-assisted pronunciation training in data-scarce scenarios.

Keywords: pronunciation correction; graph neural network; GNN; transfer learning; phoneme relationship graph.

DOI: 10.1504/IJICT.2026.155123

International Journal of Information and Communication Technology, 2026 Vol.27 No.83, pp.41 - 59

Received: 14 Apr 2026
Accepted: 21 May 2026
Published online: 27 Jul 2026 *

Title: Improving the accuracy of English pronunciation using graph neural networks and transfer learning

Keep up-to-date