Research on music generation for film based on transformer with multi-grained attention
Music generation for film is one of the current research hotspots in the field of artificial intelligence.Various deep learning algorithms for music generation have been able to create pleasing film scores.However,style contral foward genre or other parameters during generation process has been somehow neglected.In this paper,an innovative music gen-eration method is proposed,which can generates music from scratch according to the target genre.Our model,a Trans-former with multi-grained attention is introduced to model music structure,and a genre auxiliary classifier discriminator with genre classification loss and generative adversarial loss is introduced to enhance the control over genre information.The subjective and objective experimental results on the symbolic music dataset with genre information constructed in this pa-per show that the proposed method outperforms previous methods in terms of generated music quality and genre control,indicating that it is useful for automatic generation of genre-conditioned film scores.
Music GenerationGenre-conditionedGenerative Adversarial NetworkTransformerFilm Scores
万方数据
维普
