Generative Bach Music Transformer Model

An AI model trained to generative music in the style of Bach

A quick overview of what the different scripts do:

• Scrape.py: Converts MusicXML style files to tokens prepped to be fed into model
• Model.py: Transformer based model that learns to generate new tokens in the style of bach
• Converter.py: Converts generated tokens back into a MusicXML file for Musescore and other music composition applications

An example of the generated Music:

_{Generated Bach-style sheet music}

<style> img[alt="Generated Sheet music!"] { width: 600px; } </style>

How to run it yourself

While currently this model is currently trained to generated Bach style music, it can learn the compositional style of any other composer:

• First, replace the current files in the data_file with your MusicXML files of a certain composer and then run the scrape.py script(you can skip this step if you want Bach style music)
• Then run the model.py file which will train the model and generate a 200 token musicl compositon
• Finally, run the converter.py file to convert the generated tokens back into MusicXML

Future changes I want to make

• The current model is trained on a very small dataset (only about 3 pieces) so getting access to a larger variety of bach pieces would help improve the model accuracy