Slimmable NAM: Neural Amp Models with adjustable runtime computational cost

TL;DR

This paper introduces slimmable Neural Amp Models that allow real-time adjustment of model size and computational cost without retraining, facilitating flexible trade-offs between accuracy and efficiency for musicians.

Contribution

It presents a novel approach to dynamically adjustable neural models with negligible overhead, enabling real-time control in audio applications.

Findings

Models can be resized on-the-fly with minimal performance loss

Demonstrated real-time implementation in an audio plugin

Quantified performance against standard baselines

Abstract

This work demonstrates "slimmable Neural Amp Models", whose size and computational cost can be changed without additional training and with negligible computational overhead, enabling musicians to easily trade off between the accuracy and compute of the models they are using. The method's performance is quantified against commonly-used baselines, and a real-time demonstration of the model in an audio effect plug-in is developed.