liuyunhaozz / musicGeneration

generate music

musicGeneration

This repo contains the code to train a RNN mode and use this model to generate music in format midi and wav. The code has been tested with Python 3.7 on Windows 10 and Google Colab.

---

Get Started with Google Colab

Jupyter notebook is a convenient tool to display the training process and training results. In train_at_colab.ipynb, we invoke the commands to train the model and generate the ABC music file. Finally we use the IPython module to display the HTML page to display the staves and play the music.

Open the notebook train_at_colab.ipynb in Google Colab and run all cells to get you music.

---

Train

1. Put abc notation music file into folder dataset/
2. run python -r requirements.txt to install the packages.
3. run train.py with the flag below.

usage: train.py [-h] [--data DATA]

optional arguments:
  -h, --help   show this help message and exit
  --data DATA  The name of abc notation music file in dataset/

​ eg: python train.py --data music.txt

---

Generate

1. Put .abc notation music file into folder dataset
2. Put .ckpt format model into folder training_checkpoints/
3. run generate.py with the flag below

usage: generate.py [-h] [--data DATA] [--modelfile MODELFILE] [--startstr STARTSTR]
                   [--length LENGTH]

optional arguments:
  -h, --help            show this help message and exit
  --data DATA           The name of abc notation music file in dataset/
  --modelfile MODELFILE
                        model file in training_checkpoints/, eg:my_ckpt
  --startstr STARTSTR   A random string you want the music to start with
  --length LENGTH       The length of the music you want to generate

eg: python generate.py --data music.txt --model my_ckpt --startstr X --length 1000

---

Note

• The project need abc2midi and timidity to convert the music from abc notation to mid and wav. Make sure you install them on your OS before you start generating music.