Biaxial Recurrent Neural Network for Music Composition

This code implements a recurrent neural network trained to generate classical music. The model, which uses LSTM layers and draws inspiration from convolutional neural networks, learns to predict which notes will be played at each time step of a musical piece.

You can read about its design and hear examples on this blog post by Daniel Johnson.

Requirements

1. Try to run biaxia_1st_construct.sh to automatically download and install required software.
2. Run biaxia_2nd_reboot.sh or manually reboot after installing.
3. Have a test using biaxia_3rd_test.sh which can tell you if all the required things are ready or not.
4. Collect you own training MIDI file data and put them into music folder in the root folder.
5. Run python main.py and enjoy it.

#Result Folders (with ending -result) Inside XXX-result folders:

• param-x.p file means the model parameters after x times iterations.
• sample-x.mid file indicates sample music generated after x times iterations, to get quick look at how well the model is.
• x.jpg is the visualization extract from the sample-x.midi file, visaulization tool is at here .
• XXX_composition_x.mid is the music composed by the final model.

##classical music

• biaxial-ori-result: under layer setting 300,300,100,50, inside composition_0.mid,composition_2.mid``composition_3.mid``composition_8.mid are the most interesting
• biaxial-lesslstm-result: under layer setting 200,200,75,35
• biaxial-leastlstm-result: under layer setting 100,100,50,25

Michael Jackson's Rock Music

• biaxial-ori-mj-result: under layer setting 300,300,100,50
• mj_composition[x].mid is the music composed by the final model. However, only mj_composition1.mid seems interesting.

Secret Garden's Music

• biaxial-ori-secretgarden-result: under layer setting 300,300,100,50
• we don't put any output due to bad result, this should be because of lack of data