Jargon (2019). These pieces were generated using a PerformanceRnn, an LSTM-based recurrent neural network that applies language modeling to polyphonic music using a combination of note on/off, timeshift, and velocity change events trainedon various compositions of its author. The model created a generative grammar of melodies and motifs that were arranged into musical pieces. Argot is the only exception, where the first section derived from a melody inputted to the Coconet model, and the second section stems from the PerformanceRnn trained on the MAESTRO dataset.
This methodology was explored to form an aleatoric compositional practice of abstraction, where the author searches for meaning and forms of language from the outputs generated by the model. This work demonstrates how a Recurrent Neural Network model can be used as a compositional aleatoric tool.
This work demonstrates how AI can be used to develop compositional
and conceptual methodologies that explore new forms of language and
expression in composition. Moreover, it showcases how AI can be explored to create compositional methodologies for creative practices.