Pytorch Implementation of Improv RNN


This code is a pytorch implementation of the popular Improv RNN model originally implemented by the Magenta team. The model is able to generate melodies conditioned on a given chord progression.
The specific model implemented in this repository is the Chord Pitches Improv model which encodes chords as the concatenation of the following length-12 vectors:

  • a one-hot encoding of the chord root pitch class, e.g. [0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0] for a D major (or minor, etc.) chord
  • a binary vector indicating presence or absence of each pitch class, e.g. [1, 0, 0, 1, 1, 0, 0, 1, 0, 0, 1, 0] for a C7#9 chord
  • a one-hot encoding of the chord bass pitch class, which is usually the same as the chord root pitch class except in the case of “slash chords” like C/E

You can either use a pre-trained checkpoint of the model or train your own using the steps below.


Install Required Libraries

pip install -r requirements.txt

Generate a Melody Given Backing Chords

A pretrained checkpoint of the model can be found in the “checkpoints” folder. The checkpoint has been trained for 1000 epochs on the OpenEWLD dataset.

python --backing_chords "C G Am F C G F C" --output out.mid

This will generate a melody starting with a middle C over the chord progression C G Am F C G F C, where each chord lasts one bar. You can modify the backing chords as you like using the backing_chords parameter. You can define where the generated midi file should be saved with the output parameter.

An example of the generated RNN features is visualized here:

Example Generated Note Events

Train Your Own Model

Download OpenEWLD Dataset

To train the model, the OpenEWLD dataset is used. OpenEWLD is a subset of the Wikifonia Leadsheet Dataset reduced to only copyright free songs. A lead sheet is a musical score that contains a notation of the melody and the underlying chord progression of a song.
The song examples are in the compressed musicxml (*.MXL) format which can be parsed in to sequences of note events using the note-seq library.

Dataset Preparation

Extract features from musicxml files and store them in a h5 file.

python --input C:/Datasets/OpenEWLD/dataset


Track metrics using Tensorboard

python --num_epochs 1000

Track metrics using Tensorboard

tensorboard --logdir ./logs/

The curves of the loss and accuracy over the training epochs are shown in tensorboard:



View Github