Music Generator for Elderly Using Deep Learning

This research proposes a new simple music generator for elderly music instruments. This music generator can generate new music in ABC notation. It composes a new melody using LSTM to rearrange existing music bars from the original melody. The input of LSTM features is note patterns in particular music bars that are transformed and encoded as hash table using WTA technique. Through music creation processes, we set the beginning music bars and let the ending music bars to be set automatically. We, then, feed the beginning music bars to the trained model to find following music bars of the music content. Therefore, the result showed the music split into 3 parts: the beginning, the middle and the end. The experiment displayed a recreation of music pieces from the original header with a 1: 10 aspect ratio and had the same number of music bars as the original ones. The music outcomes would differ significantly from the originals by measuring the differences by comparing each a bar lines between generated and original music in the same position with the XOR function.