Research On Automatic Composing Of Chinese Genre Piano Music Based On Deep Learning

Music is one of the important forms of human emotional expression.Music genre is conceptually considered a difficult to quantify form of expression,and it undergoes rich changes with the development of music.How to use artificial intelligence methods and technologies to enable computers to learn and understand the rules in music,obtain the expression patterns of human music,and then achieve computer automatic composition has become a hotspot research direction in recent years.Compared to traditional manual composition,automatic composition can reduce the professional knowledge required for composition,allow people without any musical foundation to participate in music composition,and also provide some inspiration for professional musicians in composition.Automatic composition has important applications in generating copyrightless music,and has significant research value and practical significance in promoting the development of emotional artificial intelligence based on multi-source perception.Currently,there is no systematic research on the Chinese genre piano music in the deep learning symbolic automatic piano composition.The existing application of piano music representation in the melody centered homophonic piano music still has certain shortcomings,while a considerable portion of Chinese genre piano music belongs to the homophonic piano music.The lack and inadequacy of objective evaluation indicators also affect the development of automatic composition research.This article focuses on the representation of Chinese genre music,objective evaluation indicators of Chinese genre music,and related deep learning automatic composition models.The main work is as follows:1.Designed a MA-REMI representation based on the melody and accompaniment.Firstly,a C4-Skyline piano midi melody labeling method was proposed,and then it was combined with REMI representation.Finally,a MA-REMI representation was designed.Compared with REMI representation,the piano midi music representation sequence after MA-REMI representation has a distinction between the note of melody or accompaniment,which is more in line with the characteristics of the homophonic music.This lays the foundation for the automatic composition model to effectively learn the relationship between the melody and melody,accompaniment and accompaniment,the melody and accompaniment in Chinese genre piano music.2.Designed objective evaluation indicators for the melody’s pentatonic scale.According to the characteristics of Chinese genre music in the melody,based on the“Gong” style in the pentatonic mode,the major pentatonic scale and minor pentatonic scale in the melody of piano midi are calculated separately.From an objective perspective,evaluate whether piano midi has a Chinese genre,exclude the subjectivity of manual evaluation,and reduce related labor costs.The experimental results indicate that Chinese genre piano music composed in the “Gong” style has a higher major pentatonic scale or minor pentatonic scale in the melody compared to other genre of piano music.3.Two Chinese genre piano automatic composition models,MA-LSTM and MA-Transformer,were designed based on LSTM network and Transformer network under MA-REMI representation.And the above automatic composition model was compared and analyzed with existing piano automatic composition models Polyphony RNN and Pop music transformer in terms of model performance,objective evaluation,and artificial subjective evaluation.The experimental results show that the MA-LSTM and MA-Transformer automatic composition models can generate music of the same quality or even better than the Polyphony RNN and Pop music transformer automatic composition models with fewer parameters and faster composition time.In terms of the generation and performance of Chinese genre music,the MA-LSTM and MA-Transformer automatic composition models have better results.