Extraction of beat features for piano teaching performance based on improved autoencoder

In the process of piano teaching, accurate extraction of beat features is crucial for assessing the performance level. For the purpose of coping with the issue of insufficient characteristic extraction in the current study, this article first accelerates the convergence speed based on the asymmetric convolutional optimisation autoencoder (ACAE). Then, we obtain the note frequency of the piano performance through the start point detection algorithm, use the relationship between notes and beats to subdivide the beat detection interval, and realise the division of beats through the measurement of confidence level. Finally, the channel attention module (CAM) is introduced into ACAE to complete the adaptive weighting of beat characteristics of every channel, so as to obtain the deep beat characteristics. The experimental outcome demonstrates that the offered approach has a feature classification accuracy of 96.38% and can effectively extract beat features.