La música de los ordenadores

Abstract

[en] This final degree project has this specific name because my goal is to compose classical music with modern computers. My question is: Is it possible to compose a classical song with the actual computers and artificial intelligence? Since these last few years, the research in music generation in the classical field has been very limited. For specific cases and financed by large companies like the project of Huawei where they ended the unfinished Schubert symphony, or the DeepBach project where the final goal is to generate Bach corals (simple songs played by 4 different voices). My objective is to compose a Chopin waltz with three different neural networks. The first network will generate the harmony of the song, the second neural network will generate the accompaniment for the left hand and finally the last neural network will generate the melody of the song for the right hand. For the amount of time, it takes to get good quality music in electronic format (MIDI) I only could use 25 waltzes to train the neural networks, but it could generate accurate enough an accompaniment for the left hand from a harmony, but I have no been able to get to implement the neural network for the right hand. If you want reliable and decent results you need a lot of time and effort to collect all the scores in electronic format, so if you want to do a project of this size, dedicate all the maximum time to the collection and processing of the songs for the neural network.