Music Generator Using Deep Learning

Music Generation has become an active area of research for the past decade with dominance of Neural Networks across the AI field. Using such systems can aid musicians and composers in the process of making music. It can also be in great help for beginners that in a hunt for creating creative melodies and harmony during learning. Our project tries to use state of the art approaches in tackling this problem as a language modeling approach. Similar approaches were used in Musenet Open AI and Musicautobot. It processes midi files and converts it to text to build the language model. We then use the language model to predict and generate music. We used musicautobot, fastai and music 21 as tools to build the model. We have trained two models for classic and Pop music. We forwarded some of the model results to a musician to evaluate it and his response was that the model was capable of building simple mono-rhythm predictions and all the measures are in a correct tempo and beat stress, however, most of the endings of the predictions were bad in terms of harmony and melody.