Monophonic Piano Generation with LSTM
- python = 2.7
- pip install
tensorflow==1.14.0
keras==2.3.1
h5py
music21 - sudo apt install
ffmpeg
timidity
- The dataset is in
music_midi
. python train.py
Set training 100 epochs in default, but it can be stopped anytime (Ctrl + C).
-
Choose the best parameter
.hdf5
produced in training process
and rename itbest-weights.hdf5
-
python generate.py
Generateoutput.mid
firstly, and then turn it intooutput.mp3