A Preliminary Study On The Ai Model For Composing

With the rapid development of artificial intelligence,the use of neural network models to generate music has become an increasingly important topic in the field of music technology.Although there have been some remarkable achievements,so far,the actual application value is still relatively low.At what stage has artificial intelligence music generation developed? Which models are more advantageous? Why are most AI composition music not "human" ? What bottlenecks are encountered?What could be improved? Is there any progress in the direction of arrangement? It's a question that the author is very confused.I want to explore the answers to these questions,but I find that due to the emerging field,domestic research is relatively poor.Driven by the times,I combined my professional and arranger hobbies,hoping to make some contributions to this field.This article is divided into five chapters.Taking the background and significance of the topic as the first chapter,and the second chapter systematically expounds the basic theory and analysis angle of deep neural network learning.In the third chapter,the survey compares the eight most representative models for arrangement in this field.The method is to distribute and collect hearing score questionnaires to everyone in the survey group,make inferences and sort out,and finally get There are two models that have advantages,namely Leadsheet Generation and Music Transformer.Subsequently,in the fourth and fifth chapters,these two models were studied in depth,including further model analysis,generating experiments and summarizing conclusions and recommendations.After analyzing different models including generative music,I have put forward some suggestions for model improvements and possible future development directions in the article.I hope to help promote the actual development of this field and make up for the academic gap in this field in China.