BrainDamage is a musical instrument (VST plugin) that uses neural networks and machine learning to produce strange new sounds.
Nearly all synths recycle the same basic methods over and over: wavetable, sampling, subtractive, additive, and FM.
BrainDamage attempts to do something different; rather than start with fixed basic waveforms, you train simple neural network models to mimic a waveform, and then tweak its neurons and synapses (apply brain damage) to bend the sound in unpredictable and interesting ways.
Every neuron and synapse is effectively a unique knob you can play with, that warps sound in its own way. And every model (brain) is different, even if trained to mimic the same waveform. It rewards exploration and lends itself to happy accidents.
The world has enough great conventional synthesizers - why not try something weird?
BrainDamage is in its alpha / prototype stage; it's currently in need of some adventurous musicians and guinea pigs to explore the possibilities and help make it awesome. If you're interested, grab a copy via the download link up top, and check out the included Readme file to get started.
Honest feedback is appreciated - feel free to email me (Luke Rissacher, the developer), or leave a comment on the KVR forums.
With the current prototype you can train your own models, design sounds, save presets, and start applying brain dama%%gc.3#e to your tracks.
The fundamentals - training, sound synthesis, and neural damage - are now working, with a nice high-framerate OpenGL UI for immediate feedback and wave tweaking.
Future plans include a flexible modulation system, better MIDI control, an in-plugin model training UI, onboard effects, and possibly new neural network types specialized in producing noise-like, metallic, and inharmonic sounds.
Also, any crazy and interesting ideas testers come up with!