"Dope Learning" – Generative Music using LSTMs

Music generation using deep learning and tensorflow We implemented an LSTM over MIDI note events.

This project requires Tensorflow and the Python-MIDI library.

Installation

Run pip install -r requirements.txt to install required packages.

Running

Run python music_model.py --train <TRAINING FILES> --model_save_path <WHERE TO SAVE MODEL> -o <GENERATED MUSIC FILENAME> to train a model on the given data, and to generate a music file given the first note of the training data as context.

Summary of files

• music_model.py - File for training and running model
• preprocess.py - File for reprocessing midi data
• note_stats.py - File for analytics on MIDI tracks
• schoenberg,py - File for some more metrics on generated music data.