Drum and Melody Generation using LSTM - based Neural Networks
For this project, I constructed two LSTM - based neural networks that can generate monophonic melodies and polyphonic drum patterns. As opposed to projects which were conducted in the past, this attempt was focused on a combination of genres rather than training on only one instrument from one genre. When generating melodies, the patterns that resulted from this challenge were somewhat chaotic with some potentially inspirational exceptions. Generated drums, on the other hand, oftentimes converged to an "average of all genres" when no controlled randomness was introduced (section 7.1.5). A better imitation of the training data can certainly be achieved by using only one genre. But the involvement of several different genres led to a more unpredictable and creative outcome.
Faculteit der Sociale Wetenschappen