Research On Automatic Composition Based On Melody Structure

Since 1950 s,artificial intelligence has made some progress in automatic composition,including: the related machine learning methods based on Markov chain and artificial neural network,the knowledge base system based on music rules and music grammars,genetic evolutionary algorithms,and hierarchical structure and cellular automatetheories.These methods have imperfections,especially the structural research methods of music is very weak,the generated music does not have the overall structure,resulting in a lower degree of musical beauty.At last,the degree of automation is not high.To solve these problems of computer automatic composition and generate melody which meets requirements of the song,this paper aims at the melody structure,combined with the 19 th century Westernclassical music,ahierarchical harmony structure recursive simplified analysis method��Schenkerian Analysis,the harmony,and melody development structure of the relevant music theory knowledge,resulting in a combination of local structural and overall structural melody.First of all,this paper uses the thought of Schenkerian Analysis,the harmony theory,the mode theory andthe interval theory,and uses the production rules to generate the backbone melody basic frame.Afterward,this paper uses harmony and other music theory to transform them into the main chord.Then uses the chord mapping and other methods to turn the trunk chord in a specific note sequence and add the modified non-chord tone,then generate the theme phrase which lays the foundation for the local structure of melody.Next,through the fuzzy reasoning method for the overall control of the development of melody,this paper uses the repetitive,pinch offset,phrase-reversing and other regular designed methods to expand the theme phrase into a globally structredmelody.Finally,we build an automatic composition system based on the melody structure,and use the platform to generate a number of music.Through the manual evaluation and the machine evaluation of these songs,the results of the evaluation show that the method can output music works with structural melody and local structure and global structure and improves the aesthetic degree of the music,thus improving the degree of automation of the composition.This study improves the problem of monotonous structural weakness of computer intelligent composition,and promotes the further research on the structure of intelligent composition.