local_conv_music_generation

Music Generation with Local Connected Convolutional Neural Network.

Developed by: Zhihao Ouyang, Yihang Yin, Kun Yan.

Demo of this project is released at Demo Link.

Set Up the Environment

Required Python3 Packages:

keras
python-rtmidi
pretty-midi
progressbar

We've writen these in script, for your convenience, just do this in shell:

$ sh init.sh

Train the Monophony Model

Run the following command to launch experiment.

$ cd monophony
$ python3 monophony.py <model_id>

Replace the <model_id> to change the model for comparison.

    0 for 'conv1_model_a'
    1 for 'conv1_model_b'
    2 for 'conv1_model_c'
    3 for 'conv1_model_naive'
    4 for 'conv1_model_naive_big'
    5 for 'resnet_model_naive'
    6 for 'resNet_model_local'
    7 for 'LSTM_model'

Train the Polyphony Model

First Step

Download the training data from this link

Then extract the data into ./polyphonic/datasets

Second Step

Run the following command to launch experiment.

$ cd polyphony
$ python3 polyphony.py <model_id>

Replace the <model_id> to change the model for comparison

    0 for 'conv1_model_a'
    1 for 'conv1_model_b'
    2 for 'conv1_model_c'
    3 for 'conv1_model_naive'
    4 for 'conv1_model_naive_big'
    5 for 'resnet_model_naive'
    6 for 'resNet_model_local'
    7 for 'LSTM_model'

Contact

[email protected]

[email protected]

[email protected]