Research On Music Generation Method Based On Variable Neighborhood Search

With the rapid development of society,people’s demand for music is becoming more and more abundant,and the use of music scenes are constantly expanding.People are not only satisfied with listening to beautiful melody and enjoying the moving music,but also put forward higher requirements for modern composition.With the gradual improvement of computer computing power,researchers hope to find a way to help composers complete repeatable parts of the work,and reduce the threshold of composing,so that music lovers can enjoy the process of composing.Firstly,this paper studies the purpose and significance of music generation methods,and analyses the research status of music generation methods based on traditional methods and machine learning at home and abroad.Secondly,the application of pitch saliency based Melody Extraction and variable neighborhood search algorithm in music generation is analyzed.In addition,this paper studies a series of problems existing in traditional and machine learning music generation methods as follows:(1)According to the requirement of easy acquisition of training samples and close to the times,the input format before pretreatment is WAV or MP3,followed by mono channel extraction and segmentation.After the discrete Fourier transform,each frame is extracted based on pitch saliency,and the melody line is obtained initially.At the same time,the energy is divided into blocks based on masking effect and isophonic curve in time domain to judge the frame with notes and the frame with rest.After mapping to the note field,logarithmic weighted ambiguity processing is carried out to deliberately make the melody line appear false pitch,and then the operator’s sense of sound is exercised in the following steps.Then the music table is visualized through GUI,and the music score can be added or deleted according to preference.Then the music score is input into the training model based on music theory knowledge,and the training parameter matrix is obtained,which can be combined with the training parameters and update the training database.(2)The music generation algorithm based on variable neighborhood search is studied.According to the problem of music generation in this paper,some steps in the traditional variable neighborhood search algorithm are modified.Subsequently,because a large number of instruments are difficult to obtain,and there is a need for human ears to listen to the music score in the system,the algorithm for generating analog music is studied.The existing models of musical instruments and vocal generation are summarized and summarized.The structure of parametric musicsimulation algorithm is divided into two parts:excitation source and resonant cavity.It is realized simply by using stringed instruments and body instruments.It can add musical instruments to melody and in turn guide the structural design of musical instruments.(3)Taking reading a piece of music as an example,the training system is tested,which basically realizes the system requirements,embodies the idea of human-computer interaction,proves the validity of the secondary folding parameters,verifies the convenience and feasibility of the system,and tests and analyses the music iteration generation system.The system basically meets the requirements of convenience and flexibility of this project.The interface is intuitive,the iteration generation process is clear,the test results are also artistic,and the final output of music fragments is ideal.