A neural amp model that actually turns smoothly.
Most "parametric" NAM captures bundle a handful of fixed snapshots and switch between them at set knob positions — turn the knob and you jump from one capture to the next. ATIJ's parametric engine trains a single neural network conditioned on the knob's value alongside the audio, so it genuinely interpolates — and extrapolates beyond — the settings it was trained on.
Stepped switching vs. true interpolation
Some parametric .nam files bundle a few full captures into one file and jump between them at fixed knob positions. Ours trains a network that takes the knob position as an input alongside the audio — so it responds in between, and beyond, the settings it was actually captured at.
Typical "stepped" parametric
Turning the knob jumps from one captured setting to the next.
ATIJ's parametric engine
The response sweeps smoothly across the whole range of the knob.
How it works
Reamp
Play the same signal through a real amp or pedal at several settings of one control — e.g. gain from clean to full dime.
Train
A small neural network learns the sound at every setting in between, not just the ones that were captured.
Play
A real-time C++ engine runs the trained model in a JUCE plugin, so turning the virtual knob sweeps the tone continuously in your DAW.
Status
Coming soon as a free download. Parametric NAM builds on Steven Atkinson's open-source Neural Amp Modeler; its parametric training pipeline, model format and real-time engine are ATIJ's own work, written to make the knob a model input.