A Comprehensive Study on Algorithm-Based Music Generation

This paper is aimed to investigate the discrepancy between different methods for music generation. We claim that comparing traditional and classical music generation algorithms, music generation methods using VAEs are more favored by the listener. In this paper, we meticulously present computer-based music generation approaches including the Markov chain, genetic algorithm, and deep learning methods. The number of parameters for the deep learning method is much larger than traditional optimization algorithms, which will optimize songs that are more aesthetically pleasing. Finally, we select twenty participants to compare the music generated by different methods, the test result shows that participants prefer more on the music generated by the VAEs.