Yuanyuan Ma
The Guzheng, a traditional Chinese plucked string instrument, is known for its expressive tone, but learners and digital applications struggle with achieving accurate pitch and tonal consistency. Current manual tuning methods are labour-intensive and do not adequately preserve harmonic nuances, indicating a need for an automated tuning solution. The objective of this research is to design a deep learning (DL)–driven system capable of analysing Guzheng tones and performing real-time pitch correction, enhancing tonal accuracy and stability across various playing styles. Musical tone audio data were collected by recording high-fidelity 5000 audio samples with the IDs from multiple professional and amateur Guzheng performers under controlled acoustic conditions, capturing diverse string characteristics and playing styles. Preprocessing included spectral gating–based noise reduction to remove environmental interference and amplitude normalization to standardize signal levels. Feature extraction was performed using Mel-frequency cepstral coefficients (MFCCs) and Constant-Q Transform (CQT) to represent harmonic, timbral, and pitch-aligned spectral information. The proposed framework employs a Weighted Sparrow Search–Attention Mechanism based Temporal Convolutional Neural Network (WSS-ATT-Temp Convo-NNet), integrating temporal convolution for sequential dependencies and attention mechanisms optimized via the WSS to focus on critical tonal segments. An intelligent pitch calibration module applies real-time adaptive corrections, refining the calibration to minimize perceptual tuning errors. Experimental results under the 80%–20% split achieved MSE 0.0109, MAE 0.071, RMSE 0.104, and R² 0.972, with an inference time of 8.0 ms. The proposed approach offers a comprehensive solution for intelligent Guzheng tuning and the enhancement of traditional music performance, bridging cultural heritage and modern AI techniques.