Recognition and Error Correction of Piano Playing Music based on Spatial Attention Mechanism with Convolutional Neural Network
Jing Zhao, Kun Yu · 2024
This Nowadays, piano education industry has developed significantly which captures large share in market and reflects enhancing demand for piano learning and music education. However, automatic estimation function of piano performance has inadequacies in piano education. The Spatial Attention Mechanism with Convolutional Neural Network (SAM-CNN) is proposed for piano playing music for recognition and error correction. The SAM-CNN enhances the recognition accuracy through integrating spatial attention and focus on significant part of input data at classification which leads to better handling and complex music patterns. The CNN extracts feature like notes, rhythm patterns from preprocessed data and identifies most significant data for music recognition. The median filter was applied to smooth the input through removing noises and reserving significant data features which is helpful for accurate musical pattern recognition. The SAM-CNN performance is measured with metrics of precision, accuracy and recall. The SAM-CNN attains 97.76% precision, 98.57% accuracy and 98.12% recall for Musical Instrument Digital Interface (MIDI) dataset which is superior than Fuzzy Neural Network (FNN).