This repository contains the implementation of the DeepBach model described in
DeepBach: a Steerable Model for Bach chorales generation
Gaëtan Hadjeres, François Pachet, Frank Nielsen
ICML 2017 arXiv:1612.01010
The code uses python 3.6 together with Keras and music21 libraries. It includes a Python Flask server and a MuseScore plugin providing an interactive use of DeepBach. Examples of music generated by DeepBach are available on this website
You can download and install DeepBach's dependencies using Anaconda with the following commands:
git clone git@github.com:SonyCSL-Paris/DeepBach.git
cd DeepBach
conda env create -f environment.yml
Make sure either Theano or Tensorflow is installed.
You also need to configure properly the music editor called by music21.On Ubuntu you can eg. use MuseScore:
sudo apt install musescore
python -c 'import music21; music21.environment.set("musicxmlPath", "/usr/bin/musescore")'For usage on a headless server (no X server), just set it to a dummy command:
python -c 'import music21; music21.environment.set("musicxmlPath", "/bin/true")'usage: deepBach.py [-h] [--timesteps TIMESTEPS] [-b BATCH_SIZE_TRAIN]
[-s SAMPLES_PER_EPOCH] [--num_val_samples NUM_VAL_SAMPLES]
[-u NUM_UNITS_LSTM [NUM_UNITS_LSTM ...]] [-d NUM_DENSE]
[-n {deepbach,skip}] [-i NUM_ITERATIONS] [-t [TRAIN]]
[-p [PARALLEL]] [--overwrite] [-m [MIDI_FILE]] [-l LENGTH]
[--ext EXT] [-o [OUTPUT_FILE]] [--dataset [DATASET]]
[-r [REHARMONIZATION]]
optional arguments:
-h, --help show this help message and exit
--timesteps TIMESTEPS
model's range (default: 16)
-b BATCH_SIZE_TRAIN, --batch_size_train BATCH_SIZE_TRAIN
batch size used during training phase (default: 128)
-s SAMPLES_PER_EPOCH, --samples_per_epoch SAMPLES_PER_EPOCH
number of samples per epoch (default: 89600)
--num_val_samples NUM_VAL_SAMPLES
number of validation samples (default: 1280)
-u NUM_UNITS_LSTM [NUM_UNITS_LSTM ...], --num_units_lstm NUM_UNITS_LSTM [NUM_UNITS_LSTM ...]
number of lstm units (default: [200, 200])
-d NUM_DENSE, --num_dense NUM_DENSE
size of non recurrent hidden layers (default: 200)
-n {deepbach,skip}, --name {deepbach,skip}
model name (default: deepbach)
-i NUM_ITERATIONS, --num_iterations NUM_ITERATIONS
number of gibbs iterations (default: 20000)
-t [TRAIN], --train [TRAIN]
train models for N epochs (default: 15)
-p [PARALLEL], --parallel [PARALLEL]
number of parallel updates (default: 16)
--overwrite overwrite previously computed models
-m [MIDI_FILE], --midi_file [MIDI_FILE]
relative path to midi file
-l LENGTH, --length LENGTH
length of unconstrained generation
--ext EXT extension of model name
-o [OUTPUT_FILE], --output_file [OUTPUT_FILE]
path to output file
--dataset [DATASET] path to dataset folder
-r [REHARMONIZATION], --reharmonization [REHARMONIZATION]
reharmonization of a melody from the corpus identified
by its id
On the first run it will preprocess the dataset (it may take ~1 hour) and create a model (possibly untrained). You can download preprocessed data and pre-trained model (see below) to save some time to quick-start experimenting.
Generate a chorale of length 100 and show it in the editor:
python3 deepBach.py -l 100
In addition save it to MIDI (useful for work on a server without GUI):
python3 deepBach.py -l 100 -o output.mid
Create a DeepBach model with three stacked lstm layers of size 200, hidden layers of size 500 and train it for 10 epochs, and then sample it:
python3 deepBach.py --ext big -u 200 200 200 -d 500 -t 10
Generate chorale harmonization with soprano extracted from midi/file/path.mid using parallel Gibbs sampling with 20000 updates (total number of updates) and 16 updates per parallel batch:
python3 deepBach.py -m midi/file/path.mid -p 16 -i 20000
Use another model with custom parameters:
python3 deepBach.py --ext big -t 30 --timesteps 32 -u 512 256 -d 256 -b 16
Use another database, your dataset folder must contain .xml or .mid files with the same number of voices:
python3 deepBach.py --dataset /path/to/dataset/folder/ --ext dowland -t 30 --timesteps 32 -u 256 256 -d 256 -b 32
Reharmonization of a melody from the training or testing set:
python3 deepBach.py -p -i 40000 -r 25
Default values load pre-trained DeepBach model and generate a chorale using sequential Gibbs sampling with 20000 iterations
Put deepBachMuseScore.qml file in your MuseScore2/Plugins directory.
Run local Flask server:
export FLASK_APP=plugin_flask_server.py
flask run
or a public server (only one connection is supported for the moment).
export FLASK_APP=plugin_flask_server.py
flask run --host 0.0.0.0
Open MuseScore and activate deepBachMuseScore plugin using the Plugin manager. Open a four-part chorale. Press enter on the server address, a list of computed models should appear. Select and (re)load a model. Select a zone in the chorale and click on the compose button.
This plugin only generates C major/A minor chorales with cadences every two bars. This is a limitation of the plugin, not the model itself.
Please consider citing this work or emailing me if you use DeepBach in musical projects.
A pretrained model and pre-processed dataset is available via DropBox. Download and extract the archive contents to datasets/ and models/ in the DeepBach project root folder:
./download_pretrained_data.shMake sure DeepBach project is in your PYTHONPATH:
export PYTHONPATH=/Path/to/DeepBach/Project
music21.converter.subConverters.SubConverterException: Cannot find a valid application path for format musicxml. Specify this in your Environment by calling environment.set(None, '/path/to/application')
Either set it to MuseScore or similar (on a machine with GUI) to to a dummy command (on a server). See the installation section.