Awesome

NeuralAudio

NeuralAudio is a C++ library designed to make it easy to use neural network machine learning models (ie: guitar amplifier captures/profiles) in real-time audio applications.

Supported Models

NeuralAudio currently supports the following model types:

• Neural Amp Modeler (NAM) WaveNet and LSTM models
• RTNeural keras models (LSTM, GRU)

Underlying Libraries and Performance

RTNeural is used for RTNeural keras models.

The NAM Core implementation is used by default for NAM models, both LSTM and WaveNet.

You can also instruct the library to load NAM models using RTNeural (see the API overview below).

When using RTNeural, the official NAM WaveNet model architectures ("Standard", "Lite", "Feather", "Nano") are loaded using RTNeural by default, and use pre-compiled static architectures. Other NAM WaveNet model architectures will fall back on using the NAM Core implementation.

A subset of LSTM models are processed using pre-compiled static architectures (increasing performance). Currently the following architectures are accelerated:

• LSTM 1x8
• LSTM 1x12
• LSTM 1x16
• LSTM 1x24
• LSTM 2x8
• LSTM 2x12
• LSTM 2x16

Other architectures will work fine, but will have somewhat reduced performance.

API overview

To load a model:

NeuralModel* model = NeuralAudio::NeuralModel::CreateFromFile("<path to model file>");

To process a model:

model->Process(pointerToFloatInputData, pointerToFloatOutputData, int numSamples);

Use model->GetRecommendedInputDBAdjustment() and model->GetRecommendedOutputDBAdjustment() to obtain the ideal input and output volume level adjustments in dB.

If you would like to force the use of RTNeural for NAM models, you can.

For LSTM:

NeuralAudio::NeuralModel::SetLSTMLoadMode(NeuralAudio::PreferRTNeural);

For WaveNet:

NeuralAudio::NeuralModel::SetWaveNetLoadMode(NeuralAudio::PreferRTNeural)

Software Using NeuralAudio

The following applications are using the NeuralAudio library for model processing:

• neural-amp-modeler-lv2: LV2 plugin for using neural network machine learning amp models.
• stompbox: Guitar amplification and effects pedalboard simulation.