"Using Machine Learning in a Creative System to Parametrically Control a Modular Synthesiser"
A research project to explore the role of Neural Networks as a composer for and collaborator to the user. Informed learning trains the system which communicates with modular synthesis platforms.
With this project I aimed to design a creative method of controlling modular synthesiser parameters. Focussing on the ability of the computer to generate a vast array of textures on the synth, I then used machine learning to compose according to live user input. I chose to investigate connectivity with modular synths due to a desire to control these machines remotely with dynamic transitions. I was striving for an intelligent system that can learn and explore, exhibiting creative tendencies.
Max/MSP was used for the programming of the project. The interface is split into several windows.
Wekinator is used to perform the learning. A multilayer perceptron (MLP) is the type of neural network algorithm used to conduct the learning.
The system communicates with hardware modular synths via MIDI and an appropriate MIDI to CV interface. Software synths use internal MIDI ports or OSC.
A full blog of the project's development can be found here: