Deep Learning Algorithm Composition System Based On Music Score Recognition

With the development of computer science and music technology,algorithms have been widely studied and applied in the field of computer composition.The “computer generated art” that evolved from this belongs to the category of algorithmic art.The creator makes the computer generate music automatically or assist himself to complete music creation by writing programs and making relevant rules.Since algorithmic composition became a research hotspot in the field of computer music creation,the related theory and technology related research and exploration have been deepened continuously,and the multi-dimensional integration of science and technology and music has entered a new historical period with the rise and application of artificial intelligence,deep learning and other technologies in recent years.Artificial intelligence music composition will also become a major branch of algorithmic music composition in the future.The research purpose of this paper is to implement an algorithmic composition system based on deep learning.The theoretical research of this paper is based on the field of algorithmic composition,and the technical implementation research is based on the level of computer deep learning,as well as the corresponding ways of music data expression.The research significance is to obtain the output results with strong music data characteristics in a short time and effectively balance the data volume,calculation cost and time cost.At the same time,the complexity of the deep neural network can be adjusted to determine the pros and cons of the algorithm again,so as to continuously upgrade and optimize the system.In this paper,the research content includes the theory background,application feature analysis of the common algorithms,the building process of OMR system,deep learning training strategies,music data expression,MIDI data standardization,music data preprocessing methods,deep learning neural network building,music generated fragment analysis and so on,the research methods include: literature study method,experimental study method,interdisciplinary study method,comparative study method and exploratory study method,etc.In this paper,how to use the supervised machine learning method to produce music with polyphonic structure is discussed.The results of this study are constructing three system modules: OMR score recognition system,data training system and music generation system as well as comprehensively building a set of data sampling tools synthesized by OMR system.The research conclusion of this paper is based on LSTM network,BiLSTM network and deep learning network under attention mechanism,which can more effectively extract features in music data.The results of this study can basically meet the composer's auxiliary creation needs,and music researchers' needs for analysis of a large number of music data.The results can also be transplanted to music deep learning research,human-computer interaction music creation and applied music creation for further expansion.