Double-track Music Generation Based On GAN Network With Chord Constraint

Since Generative Adversarial Network(GAN)was proposed,there have been a lot of research achievements in speech,text and visual generation,but little research in music generation.Secondly,chord in music can make music more melodious,has the function of restraining the generated music and increasing the fullness.However,there are few studies on the application of chord to the model in the field of music generation.Finally,music is a kind of temporal art,which progresses with time,so the dependence of notes with time should be added to the model when constructing the model.Therefore,this paper studies the music generation on the basis of the existingtheory,technology and research.We take Deep Convolution Generative Advers arial Network(Midinet)as the baseline model.We constructed three models,w hich are respectively the music generation model based on music theory rules,deep chord convolutional generative adversarial network(DCC?GAN)and deep chord gate recurrent unit generative adversarial network(DCG?GAN).The thre e models were used to learn music theory rules,extract chord features and con struct the overall style respectively,which ultimately made the generated music more harmonious and melodious.The music generated by this model is different from the traditional one in that it does not need to add complex music rules manually,but directly genera tes music with music theory rules by training the initial music dataset,and im proves the stability of generated music through the constraint of chord,and in crease the time dependence of the notes.In the analysis of the experimental res ults,50 people(40 normal people and 10 music professionals)were invited to e valuate and analyze the generated melodies.Deep Chord Gate Recurrent Unit Generative Adversarial Network proposed in this study is more pleasing throug h chord constraint and self-learning content of the moment 1— t-1,which has important theoretical and practical significance for the creation and realization of popular music and mass production.