Here, we attempt to dissect Open AI Scholar Christine Payne's project on Musical Neural Net and further her work to complete some unfinished compositions by famous musicians.
These instructions will get you a copy of the project up and running on your local machine.
-
Clone this Github repo into your local machine.
-
You will need to have the following packages install: pytorch, music21, jupyter, numpy
-
Open the iPython Notebooks and follow the tutorial inside.
Notewise text files of piano, jazz, and chamber (piano & violin) can be downloaded here:
-
midi files for training goes into midi-files
-
encoder takes input midi files and converts them to text files into txt-files/notewise/note_range62/sample_freq12/
-
word_model takes text files as input and outputs generated text files into txt-files/notewise/custom
-
notewise-decoder.py takes text files and decodes them back to midi files into output-midi-files
notewise_decoder.py
a script to decode any generated txt file into midi
word_model.py
our baseline 2-layer lstm model (for cpu)
notebooks/2018-11-14_model02-cuda.ipynb
latest version of our 2-layer lstm model (for gpu)
experimenting with hyperparameters
output-midi-files/notewise/custom/run01_2018-11-14/
contains samples of generated files during training
- Christine Payne, Open AI Scholar, for her amazing work!
- DJ Gan Team @ Berkeley MEng
If you face any issues or have any questions, please contact me at edwardtky@berkeley.edu