Published Paper Details:

Artificial intelligence solutions have been brought into the creative field such as generations of art and music. Generating music via deep learning is a challenging problem, however recent developments such as openAI's MuseNet, CTRL (Creator Technology Research Lab) at Spotify, and Google's Magenta have shown us that there can be viable solutions. However, these technologies use very complex models trained over thousands of samples of data. This paper presents a survey on different models used in the past and also proposes a new model which uses LSTMs (Long Short Term Memory) and a small dataset to show that a relatively basic model is also functional in producing a sequence of notes which is capable of demonstrating properties resembling a nice sounding piece of music. We also show that a model produces satisfactory results when presented with a small dataset to train. This paper also discusses the challenges that accompany and discusses the strategy, and architecture that is involved in creating a system that can successfully generate a piece of music rather than a sequence of notes.