This directory contains scripts and other files needed to prepare the training and test datasets. Before starting, make sure that the groove2groove package is installed in your environment.

The scripts will download the data and perform some preprocessing. This includes conversion to the format used by Groove2Groove: an LMDB database of Magenta note sequences plus a JSON file with metadata.

1. Go to the synth directory and run ./download.sh. This will download the synthetic dataset from Zenodo and extract the relevant files to the train, val and test subdirectories.
2. Preprocess each part of the dataset:

   ./prepare.sh train
   ./prepare.sh val
   ./prepare.sh test
   

Go to the bodhidharma directory and run ./prepare.sh. This will download, extract and preprocess the dataset.

Additionally, velocity normalization needs to be performed by running the vel_norm.ipynb notebook.