The SmartAmp features two deep learning models using clean/lead channels, and the ability to interchange new models with the Load Tone button. Each channel has separate bass, mid, treble, and gain settings, with a presence knob for overall sound.
The SmartAmp uses a modified version of the deep learning neural network "WaveNet", which was developed in 2016 by DeepMind and published in the paper “Wavenet: A Generative Model for Raw Audio” (1). This model has applications for natural sounding speech generation. The WaveNet model was then modified for the purpose of guitar amp and pedal emulation by the Acoustics Laboratory at Aalto University in Finland. They published a paper in 2019 titled “Real-Time Guitar Amplifier Emulation with Deep Learning” (2), along with a real time c++ implementation on GitHub under the Apache 2.0 license. This same inference model is used in the SmartAmp for emulation of specific amps and pedals.
The WaveNet model consists of layers of parameters (weights and biases) that are trained on audio data from a guitar. The input/output audio is fed through the network, which gradually adjusts these parameters to behave like the target amp or pedal. The goal is to reduce the error to signal ratio of the model in order to replicate the hardware behavior. This approach is called "black-box" modeling, because we don't care what's going on inside the amp/pedal circuit; it only matters what the signal looks like going in and coming out. The opposite approach is white-box modeling, where individual components like tubes, resistors, and capacitors are modeled. A mix of the two approches is called grey-box modeling.
Using artificial intelligence to simulate audio hardware is a game changer in the industry. There is much to learn and much to improve upon. A known shortcoming of the WaveNet model is high cpu usage, at least when compared to other guitar plugins. The advantage here is being able to create near perfect digital copies of real-world hardware simply from audio recordings. As machine learning algorithms improve and computers become even faster, I expect A.I. will become more and more common for digital audio processing.
1. Aaron van den Oord et al., “Wavenet: A Generative Model for Raw Audio,” arXiv Preprint arXiv:1609.03499, 2016.
2. Alec Wright et al., “Real-Time Guitar Amplifier Emulation with Deep Learning,” Applied Sciences 10, no. 3 (2020): 766.
Copy the SmartAmp.vst3 file to the proper location in your DAW's plugin path.Mac Installation
VST3: Download and open the .dmg disk image. Drag the "SmartAmp.vst3" to the VST3 folder, or copy to desired directory. (by default installs to "/Library/Audio/Plug-Ins/VST3" ; may need to reboot for DAW to recognize the new plugin)
AU: Download and open the .dmg disk image. Drag the "SmartAmp.component" to the Components folder, or copy to desired directory. (by default installs to "/Library/Audio/Plug-Ins/Components" ; may need to reboot for DAW to recognize the new plugin)Ubuntu Linux Installation
Extract the "SmartAmp.vst.zip" folder to your VST3 location (for example "~/.vst3/") for your DAW to recognize. Currently has been tested using Reaper DAW for Linux). This is an experimental build of the SmartAmp.The TonePack
The TonePack contains .json models of specific guitar tones made by the GuitarML community. You can load these into the SmartAmp for an unlimited range of sounds. Load a .json file from the TonePack with the "Load Tone" button in SmartAmp. The current channel's tone/gain knobs will be applied to the custom tone. Switching the clean/lead channel will reload the default tone for the channel and unload the custom tone.