A demo application that processes audio files using slimmable Neural Amp Models (NAMs). This demo integrates NeuralAmpModelerCore with SlimmableNamDsp to allow runtime adjustment of model complexity.

• Load and process WAV audio files (mono, 24-bit, 48kHz)
• Load slimmable NAM model files (.nam)
• Automatically detects all available slimmable sizes
• Processes audio with each available size
• Saves separate output files for each size (e.g., output-1.wav, output-2.wav, ...)

• CMake 3.15 or higher
• C++17 compatible compiler
• Eigen3 library
• libsndfile library

brew install eigen libsndfile cmake

sudo apt-get install libeigen3-dev libsndfile1-dev cmake build-essential

vcpkg install eigen3 libsndfile

Clone the repository (including submodules):

git clone --recursive <repository-url>

Or if you've already cloned:

git submodule update --init --recursive

Build the project:

./build.sh

Run with default files:

./build/slimmable-demo

This will process data/pop-punk.wav with data/slimmable-model.nam and create output/output-1.wav, output/output-2.wav, etc.

Using the build script (recommended):

./build.sh

Manual build:

# Make sure submodules are initialized
git submodule update --init --recursive

mkdir build
cd build
cmake ..
make

Windows with Visual Studio:

mkdir build
cd build
cmake .. -G "Visual Studio 17 2022"
cmake --build . --config Release

./slimmable-demo [input.wav] [model.nam] [output_prefix]

• input.wav - Input audio file (WAV format, must be mono, 24-bit, 48kHz)
  • Default: data/pop-punk.wav
• model.nam - Slimmable NAM model file (must be a slimmable model)
  • Default: data/slimmable-model.nam
• output_prefix - Prefix for output files
  • Default: output

• Input: Must be mono (1 channel), 24-bit, 48kHz WAV file
• Output: Always saved as mono, 24-bit, 48kHz WAV file
• The application will reject input files that don't meet these requirements
• The application requires a slimmable NAM model (non-slimmable models will be rejected)

The application processes the audio with all available slimmable sizes and creates separate output files in the output/ directory:

• output/output-1.wav - Processed with size 1 (smallest/fastest)
• output/output-2.wav - Processed with size 2
• output/output-3.wav - Processed with size 3
• ...
• output/output-N.wav - Processed with size N (largest/best quality)

The output/ directory is created automatically if it doesn't exist.

Run with defaults:

./build/slimmable-demo

This uses data/pop-punk.wav, data/slimmable-model.nam, and creates output/output-*.wav files.

Specify custom files:

./build/slimmable-demo input.wav my_model.nam output

This will:

1. Load input.wav
2. Load my_model.nam as a slimmable model
3. Detect all available slimmable sizes
4. Process the audio with each available size
5. Save results to output/output-1.wav, output/output-2.wav, output/output-3.wav, etc.

• NeuralAmpModelerCore
• SlimmableNamDsp
• Slimmable NAM Paper