A Study of Music Composition Models Combining Generative Adversarial Networks and Their Application to Film Scoring

Abstract

Automatic generation of movie music is one of the current research hotspots in the field of artificial intelligence. This study proposes a music composition model based on the application of artificial intelligence in movie soundtrack composition. The model is based on generative adversarial network and its derivative model CT-GAN, and builds the MTC-GAN music generation method through multi-track correlation modeling, temporal structure modeling and discretization processing. In the objective index comparison experiments, the experimental indexes verify that the MTC-GAN model in this paper can create high-quality music, and its Scale Consistency, Tone Span, and Uniqueness structures are 89.13%, 13.84, and 64.43, respectively, which effectively improves the quality of music generated by the music composition model. The model-generated movie soundtrack score has an overall improvement of 1.48% over the human work score, and its evaluation scores in terms of melody, rhythm and emotion are optimal among all compared models. Experiments show that the method in this paper achieves ideal results in generating music quality, which is helpful to help movie soundtrack creation and promote the innovation and development of movie soundtrack art.