Automated Music Generation using Deep Learning
Shikhar Bhardwaj, Shalmiya Mundeth Salim, Talha Ali Khan, Soudeh JavadiMasoudian
Abstract
Generating music using Deep Neural Networks has been an active field of study for the past few decades. This paper discusses how to create computer-based monophonic musical content using a Recurrent Neural Network with the extension of Long Short-Term Memory. The objective of this study is to generate music that has a melodic nature to it and can be played by humans using sheet notation. The neural network is fed with input from the video game series: Final Fantasy for creating the melody. The selection of instruments is piano, which will be the same for the input and output. The main focus of this paper is on constrained music, differentiating notes from chords to help the architecture learn to distinguish one from the other and generate music with a mix of both, thus creating a pleasant listening experience.