NeuralAmpModelerCore Documentation

Welcome to the NeuralAmpModelerCore documentation. This library provides a core C++ DSP implementation for Neural Amp Modeler plugins.

Contents:

Overview

NeuralAmpModelerCore is a high-performance C++ library for running neural network-based audio processing models. It supports multiple architectures including:

  • WaveNet: Dilated convolutional neural networks with gating and conditioning

  • ConvNet: Convolutional neural networks with batch normalization

  • LSTM: Long Short-Term Memory networks

  • Linear: Simple linear models (impulse responses)

The library is designed for real-time audio processing with a focus on:

  • Real-time safety: Pre-allocated buffers and no dynamic allocations during processing

  • Performance: Optimized implementations using Eigen for linear algebra

  • Flexibility: Support for various activation functions, gating modes, and conditioning mechanisms

Getting Started

For an example of how to use this library, see the NeuralAmpModelerPlugin repository.

Architecture

The library is organized into several namespaces:

  • nam::wavenet: WaveNet architecture implementation

  • nam::convnet: ConvNet architecture implementation

  • nam::lstm: LSTM architecture implementation

  • nam::activations: Activation function implementations

  • nam::gating_activations: Gating and blending activation functions

Key Components

  • DSP: Base class for all DSP models

  • WaveNet: Main WaveNet model class

  • Conv1D: Dilated 1D convolution implementation

  • FiLM: Feature-wise Linear Modulation

Documentation

Indices and tables