An Improved Time Series Network Model Based on Multitrack Music Generation

Abstract

Deep learning architecture has become a cutting-edge method for automatic music generation, but there are still problems such as loss of music style and music structure. This paper presents an improved network structure of time series model based on multi-track music. A context generator is added to the traditional architecture. The context generator is responsible for generating cross-track contextual music features between tracks. The purpose is to better generate single-track and multi-track music features and tunes in time and space. A modified mapping model was further added to further modify the prediction results. Experiments show that compared with traditional methods, the proposed will partially improve the objective music evaluation index results.