Getting Started

Neural Network

A hidden size of 70 and 2 layers yield good results:

python train_hannds.py --hidden_size 70 --layers 2 --length 100 --cuda --cv_partition 1 --network 88 --rnn_type LSTM

Provide the --cuda flag to use the GPU.

Kalman filter

python kalman_mapper.py

Hannds server

python hannds-server.py

Then open simple-text.maxpat in Max/MSP.

Offline Processing

python single_file.py --input midi_file.mid