Research on an automatic transcription method for piano music based on artificial neural networks

Automatic transcription of music is a key process in music signal processing, which can be described as the process of transcribing raw music information into musical symbol representations. The current automatic music transcription methods don’t perform well in capturing temporal details of music, which will affect the transcription result. In order to solve the problem, this article constructed an artificial neural network model including convolutional neural networks and temporal networks to predict the onsets, offsets, and pitches of notes in piano music videos, and tested the performance of the model in the piano music dataset MAESTRO. The results of experiments showed that the model achieved good transcription performance, with the optimal model achieving an accuracy of 86% and a recall rate of 86%.