Improving the Interactive Accuracy of Intelligent Piano Teaching Using a PSO-SVM Optimized Fingering Recognition Model

Current intelligent piano teaching systems are limited by insufficient feature extraction and poorly optimized classification parameters, which reduces fingering-recognition accuracy and weakens human–machine interaction feedback. This paper proposes a high-precision fingering recognition method based on Particle Swarm Optimization and Support Vector Machine. Keystroke timing, force, and duration are first collected and normalized as discriminative feature vectors. A standard Particle Swarm Optimization algorithm is then designed with cross-validation accuracy as the fitness function to iteratively optimize the SVM kernel parameter γ and penalty factor C. The optimized classifier is trained on fingering samples and embedded into an intelligent piano teaching system for real-time recognition. Experiments on the PIG dataset, including 13,800 annotated fingering samples from ten trained pianists performing six graded piano pieces on a Yamaha U1 acoustic piano, show 91.2% recognition accuracy for two-finger fingering and an 87.0% F1-score for inter-finger repeated fingering. With an average response time below 67 ms, the system demonstrates high recognition accuracy and real-time interaction capability, providing a feasible technical path for intelligent piano instruction and short-duration motion-signal classification.