Performance RNN: Generating Music with Expressive Timing and Dynamics

Pytorch implementation of Performance RNN

0. Requirments

• Pretty_midi
• numpy
• pytorch >= 1.0

1. Usage

# Data Tree
config.data_dir/
└── config.data_name/
        ├── midi/ *.mid
        ├── note/ *.data  # processed 
        └── event/ *.data # processed
  
# Project Tree
PRNN
├── PRNN_src/
│       ├── data/ *.py
│       ├── loss/ *.py
│       ├── model/ *.py
│       └── *.py
└── PRNN_exp/
         ├── log/
         ├── model/
         └── save/         

1.1 Train

# Note based
python train.py --data_type "note"

# Event based
python train.py --data_type "event"

1.2 Test

# Note based
python train.py --is_train false --data_type "note"

# Event based
python train.py --is_train false --data_type "event"