End-to-end music score note identification method based on deep learning

Abstract

The invention discloses an end-to-end music score note identification method based on deep learning. The method is divided into three steps in total: (1) data preprocessing: downloading a corresponding data set from MuseScore, and recoding pitch and value labels; and (2) data enhancement: performing data enhancement on the recoded music score data; the invention providing four different enhancement methods; and (3) an end-to-end model: using a deep convolutional neural network model for end-to-end music score note identification, inputting the enhanced data into the model, and the output of the model being the note value and pitch. The music score note recognition model based on deep learning is provided for the printed music score, namely, the whole music score image is input into the model, the value and pitch of notes on the music score are directly output, the model is completely end-to-end, and the multi-voicemusic score image can be accurately identified.

Description

technical field [0001] The invention belongs to the field of optical music score recognition, and is an end-to-end neural network recognition method based on deep learning, which can be applied to musical note recognition. Background technique [0002] Optical Score Recognition is the musical application of Optical Character Recognition to recognize musical scores into editable or playable forms such as MIDI (for playback) and MusicXML (for page layout). Compared with other symbols in the music score, the musical note occupies a very high proportion. It is used to record the pitch and time value, and has important semantic information. Therefore, note recognition is the core and key of score recognition. The shape of musical notes is ever-changing, and its diversity and polymorphism characteristics determine that musical notes are difficult to identify. The traditional note recognition method needs to delete the staff in advance, then extract the primitive symbols, and com...

AI Technical Summary

Problems solved by technology

However, the current target detection method based on deep learning cannot recognize the pitch and duration of notes, and the sequence recognition method has problems such as low recognition accuracy when processing multi-voice scores.

Images