Title: Research on optical music recognition based on improved CRNN network and its application in piano teaching

Authors: Jianing Wang

Addresses: School of Humanities and Management, Xi'an Traffic Engineering University, Xi'an, 710000, China

Abstract: The traditional optical music recognition method has the problem of low recognition accuracy and efficiency. An optical music recognition method based on convolutional recurrent neural network (CRNN) is proposed. Firstly, residual depthwise separable convolution is introduced into convolutional layer of CRNN network. Then, after convolution operation, squeeze-excitation module in attention mechanism is introduced. Finally, parameters of cross entropy function are adjusted at transcription layer. The results reveal that error rate of note recognition and sequence recognition in optical music is 1.26% and 7.31% respectively, which is significantly lower than those of CRNN model and SE-bi-directional long short-term memory (SE-BiLSTM) model. This model can improve training speed, and its recognition time is only 6.44 s, which is 7.89 s and 14.65 s lower than that of other two methods, respectively. It shows that recognition efficiency of the proposed model is significantly improved, which can meet the actual teaching needs of piano classrooms.

Keywords: CRNN network; optical music recognition; SE module; note feature extraction; piano teaching.

DOI: 10.1504/IJCSM.2025.149900

International Journal of Computing Science and Mathematics, 2025 Vol.22 No.2, pp.176 - 191

Received: 21 Aug 2024
Accepted: 05 Jul 2025
Published online: 17 Nov 2025 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article