Children with cleft palate often experience persistent speech difficulties caused by abnormal oral structures and impaired velopharyngeal function. Conventional pronunciation training relies on repeated practice and professional feedback, yet children practicing independently can hardly judge whether their articulatory movements are appropriate. To address this problem, this study proposes a soft-palatal training system that combines a hardware training device with an artificial intelligence-based visual feedback module. The hardware part is a silicone oral appliance driven by fishing lines and pneumatic airbags, which can bend bidirectionally to guide the soft palate into a correct position. The software part extracts facial landmarks from speech videos and computes geometric, temporal, and statistical features to characterize mouth movements. A hybrid one-dimensional convolutional neural network, bidirectional gated recurrent unit, and Transformer architecture is developed to classify four predefined visual pronunciation patterns. The GRID audiovisual speech corpus is used to develop and evaluate the proposed model. The model achieves an accuracy of 87.30%, a macro F1-score of 87.43%, a precision of 87.38%, and a recall of 87.49%. It outperforms the LSTM, GRU, Transformer, CNN, and MLP baselines. These findings support the feasibility of the proposed visual feedback approach for pronunciation analysis and its potential use in home-based articulation training for children with cleft palate.

Yiyan Pei. Soft Palatal Trainer with Visual Feedback for Children with Cleft Palate. Frontiers in Medical Science Research (2026), Vol. 8, Issue 5: 41-49. https://doi.org/10.25236/FMSR.2026.080505.

[1] E. B. Stallings, J. L. Isenburg, R. E. Rutkowski, et al., "National Population-Based Estimates for Major Birth Defects, 2016-2020," Birth Defects Research, vol. 116, no. 1, p. e2301, 2024.

[2] A. W. Kummer, "Speech Therapy for Errors Secondary to Cleft Palate and Velopharyngeal Dysfunction," Seminars in Speech and Language, vol. 32, no. 2, pp. 191-198, 2011.

[3] G. Pu and H. Wang, "Review on Research Progress of Machine Lip Reading," The Visual Computer, vol. 39, pp. 3041-3057, 2023.

[4] T. Stafylakis and G. Tzimiropoulos, "Combining Residual Networks with LSTMs for Lipreading," in Proceedings of INTERSPEECH, pp. 3652-3656, 2017.

[5] S. Petridis, T. Stafylakis, P. Ma, F. Cai, G. Tzimiropoulos, and M. Pantic, "End-to-End Audiovisual Speech Recognition," in Proceedings of the IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pp. 6548-6552, 2018.

[6] A. Vaswani, N. Shazeer, N. Parmar, et al., "Attention Is All You Need," in Advances in Neural Information Processing Systems, vol. 30, pp. 5998-6008, 2017.

[7] M. Cooke, J. Barker, S. Cunningham, and X. Shao, "An Audio-Visual Corpus for Speech Perception and Automatic Speech Recognition," Journal of the Acoustical Society of America, vol. 120, no. 5, pp. 2421-2424, 2006.

[8] Y. Kartynnik, A. Ablavatski, I. Grishchenko, and M. Grundmann, "Real-Time Facial Surface Geometry from Monocular Video on Mobile GPUs," in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops, 2019.