Music-Generation-using-LSTM

A comparison between LSTM and GRU models for generating Piano Music From scratch. Dataset-> 92 piano MIDI files. As per our comparison, LSTM generated more accurate and better sounding results. (Measurement instrument being the Human Ear) These models were chosen against simple RNNs because of the vanishing gradient problem, which basically means that Recurrent neural networks tend to forget the past data and results as the loop over newer data.