We present Performance RNN, an LSTM-based recurrent neural network designed to model polyphonic music with expressive timing and dynamics. Here’s an example...