Content: This repository contains a recurrent neural network (rnn) that was trained on Chopin's nocturnes. It also contains a program that parses midi files into strings of characters that are more easily understood by the rnn.
Goal: The goal of this project was to see if a recurrent neural network could go further than just creating music but also imitate the style of an artist. Chopin's nocturnes were selected because of their specific style that can be recognized easily and because of their use of a single instrument.
Results: Although the resulting tracks showed some musicality, they lacked a sense of rhythm. In retrospection, I realize that the program I created to parse the midi files presented a loss of information that was too important, especially with the rhythm.
Conclusions: This project showed that a basic rnn could learn some musical patterns. However, it didn't seem to quite match the style of Chopin's nocturnes. Because of all the limitations of the rnn, this project would have been more interesting with a different musical style, like jazz.