RNN Music Composer

Use TensorFlow to Generate Music with a Restricted Boltzmann Machine.

Restricted Boltzman Machine

A Restricted Boltzmann Machine (RBM) is a generative stochastic Neural Network that can learn a probability distribution over its set of inputs. This setup can be used to predict the next note in a sequence given the previous note.

For more info go here

Dependencies

• Tensorflow
• pandas
• numpy
• msgpack-python
• glob2
• tqdm
• python-midi

Basic Usage

To train the model and create music, simply clone this directory and run

python gen_music.py

The run usually takes 5 minutes in a modern PC.

Changeable Parameters

• To change input midi files, modify line 28:

songs = get_songs('elvis')  #Change this to your folder name

• To change song length generated, modify line 39:

num_timesteps = 150  # number of timesteps that we will create, (change this to change the time of music generated)

• To change number of epochs, modify line 43:

num_epochs = 1000  # The number of training epochs that we are going to run.

• To change save location, modify line 135:

midi_manipulation.noteStateMatrixToMidi(S, "out/generated_chord_{}".format(i))

G00D LUCK

For doubts email me at: atinsaki@gmail.com