deep-learning deep-sequential-modelling recurrent-neural-networks rnn tensorflow long-short-term-memory lstm keras
The songs are represented in ABC notation. The model receives one or two notes to start with, then predicts the next notes to form the rest of the song. The generated song is then converted to MIDI, for you to import into your DAW or play right away. The result may or may not be impressive - if not, then try again, or train the model yourself.
Below are some of the songs the AI created, which sound pretty realistic. The first one is my favorite.
ยฉ 2020 Moses Odhiambo