Integrating Neural Amp Modeling into Csound

Abstract

Neural Amp Modeler (NAM) is the most widely used open-source system for the black-box modeling of nonlinear audio hardware using neural networks. A community library of several hundred thousand pre-trained captures exists, but there has been no way to use them inside Csound. This paper presents NAMProcess, a Csound opcode that performs real-time NAM inference. The opcode loads standard .nam model files, switches between models without interrupting audio, adds no algorithmic latency, and runs comfortably in real time on systems like the Raspberry Pi. A live dry-referenced volume compensation system built into the opcode holds every capture in a heterogeneous 25-model test library within ±2 dB of the player's volume. This paper details the implementation of a thin wrapper around the NeuralAmpModelerCore C++ library and its integration as a native plugin opcode, a drag-and-drop Cabbage plugin, and "tone selection" in a Raspberry Pi-based guitar pedal.

Keywords: Neural Amp Modeling, Raspberry Pi, Cabbage plugin, opcode, volume compensation