Article: A method for generating personalised feedback of violin performance based on conditional generative adversarial network Journal: International Journal of Reasoning-based Intelligent Systems (IJRIS) 2026 Vol.18 No.22 pp.50 - 70 Abstract: In this study, a feedback generation method based on conditional generation countermeasure network is proposed. The performance audio of 30 students with different levels (44,100 Hz sampling rate) and the corresponding teacher comments were collected, and a dataset of 465 effective segments was constructed. The generator inputs audio frequency spectrum (128 × 128) and performance level label (one-hot code), and outputs natural language feedback sentences. The discriminator judges the matching degree between the feedback and the performance content and conditions. After 200 rounds of training, the learning rate of the generator is 0.0002. CGAN's FID score is 18.6, and the user's score is 4.2 (out of 5). The feedback accuracy from the primary level to the advanced level is 89%, 86% and 81% respectively. Ablation experiment proved that the condition information reduced FID by 13.9. The system provides a quantitative and personalised technical path for intelligent violin teaching. Inderscience Publishers - linking academia, business and industry through research

Title: A method for generating personalised feedback of violin performance based on conditional generative adversarial network

Authors: Wanshu Jiang

Addresses: Department of Music Education, Zhejiang Conservatory of Music, Hangzhou, 310024, China

Abstract: In this study, a feedback generation method based on conditional generation countermeasure network is proposed. The performance audio of 30 students with different levels (44,100 Hz sampling rate) and the corresponding teacher comments were collected, and a dataset of 465 effective segments was constructed. The generator inputs audio frequency spectrum (128 × 128) and performance level label (one-hot code), and outputs natural language feedback sentences. The discriminator judges the matching degree between the feedback and the performance content and conditions. After 200 rounds of training, the learning rate of the generator is 0.0002. CGAN's FID score is 18.6, and the user's score is 4.2 (out of 5). The feedback accuracy from the primary level to the advanced level is 89%, 86% and 81% respectively. Ablation experiment proved that the condition information reduced FID by 13.9. The system provides a quantitative and personalised technical path for intelligent violin teaching.

Keywords: conditional generative adversarial network; violin playing; personalised feedback; audio analysis.

DOI: 10.1504/IJRIS.2026.155996

International Journal of Reasoning-based Intelligent Systems, 2026 Vol.18 No.22, pp.50 - 70

Received: 29 Apr 2026
Accepted: 30 Jun 2026
Published online: 28 Aug 2026 *

Title: A method for generating personalised feedback of violin performance based on conditional generative adversarial network

Keep up-to-date