[Free] Introducing ParametricOD, which reproduces pedal effects in detail using NAM's parametric modeling function.

Introducing ParametricOD, which reproduces pedal effects in detail using NAM's parametric modeling function.

ParametricOD is a plug-in that uses the previously introduced NeuralAmpModelerPlugin technology (Neural Amp Model (NAM)) to reproduce the knob and switch movements of an overdrive pedal effect.

The main features are as follows.

1. Beyond snapshot modeling: ParametricOD can accurately emulate the dynamic changes when moving physical gear knobs and switches. This challenges the common perception that neural network-based methods* are limited to static "snapshot" models.

2. Streamline data collection: Addresses the "curse of dimensionality" problem in creating multiparameter models and significantly reduces the amount of data required.

3. CPU efficiency: ParametricOD has lower CPU usage than many snapshot models, making it more accessible to users without high-performance computing resources.

4. Customizability and open source framework: We leverage the open source nature of NAM to customize the model architecture for pedal modeling. This is of particular interest to developers and audio engineers interested in exploring and extending the boundaries of digital audio processing.

5. Inspiration and resources for the community: By releasing ParametricOD as a free plugin and welcoming interaction with people interested in building their own parametric models, we are helping to form a community that fosters technological advancement in guitar effects.

*Neural DSP's Neural Capture, TONEX's Tone Modeling, Headrush's Smart amp/pedal cloning, Tonocracy's ToneSnap, etc.

Sound demo

At first, I turn the effects off and then turn them on.

afterkHs ConvolverI turned it on later and smoothed it out.

Although this is not free, it contains quite a variety of IRs and is recommended.

Overview

Today we're releasing ParametricOD, a plug-in that uses NAM's parametric modeling capabilities to provide an accurate model of an overdrive pedal across the entire range of the pedal's knobs and switches.

This plugin has similar compatibility with the open source snapshot plugin (“NAM plugin”, or maybe just “NAM” for many users), and supports VST3 and AU formats for macOS. Available in VST3 for Windows.

This plugin is intended as a "concept" plugin in the sense that we want to use it to address potential misconceptions and demonstrate existing features of NAM that may not be known to many people. :

NAM is not just a "snapshot" modeler.
Since NAM first appeared in 2019, a number of data-driven modeling products have emerged in the guitar space. Many of them (Kemper's Profiling, Neural DSP's Neural Capture, TONEX's Tone Modeling, Headrush's Smart amp/pedal cloning, Tonocracy's ToneSnap) focus on emulating the tone of your gear in a single “snapshot.” , leading to the impression that neural methods are incapable of modeling the effects of moving actual gear knobs and switches.

However, this is not accurate. Those familiar with NAM may recall that I used it to create a 7-knob parametric model:

Also, NAM is not the only project announcing this feature. For example, GuitarML's Proteus supports “knob capture”. However, I hope that this plugin will help raise awareness of NAM and help more people learn about its potential.

Parametric Modeling Doesn't Require Impossible Amounts of Data A related misconception is that it is virtually impossible to collect enough data to create such models. For models with one knob (e.g. the “drive” knob), you can imagine sweeping the knob from 1 to 0 in 10 increments. Using a standard reamp file that I have prepared for NAM, you should be able to do this in less than an hour. But to do this with two knobs you would have to do all the combinations of knobs, and you might imagine that you would need 1 x 1 = 2 reamps. For this model with 11 knobs and 11 switches*, this logic means I've done almost 121 reamps and recorded over 2 hours of audio.

One way to avoid this is to reduce the number of points. You can reduce the number of points by about 1 times in total by changing it to 1 increments (2, 0,2,4,6,8,10, 4, 1, 2, 7) instead of 100 increment. If there are only two values ​​per knob (minimum, maximum), even the XNUMX-knob model above would require over XNUMX reampings (the accuracy of interpolating between extreme values ​​may be quite questionable) (I don't know!). This challenge has a name: "Curse of Dimensions."

This is such a big problem that a lot of research is being done to solve it, much of it falling into the scientific field of optimal experimental design. The details are beyond the scope of this blog post, but by using advanced techniques in this field, I was able to reduce my time spent (including the time spent moving knobs between reamps) to just over an hour. is. Smarter, less difficult!

NAM is not inherently CPU intensive
Users should notice that the CPU load on this plugin is much lower than they experience with many snapshot models. This is achieved by using a lighter neural network architecture to save CPU while achieving "NAM-level" accuracy.

NAM can be customized
It's very difficult to point to something and say "NAM can't do that." It's not just snapshot modeling; it's purposefully designed to solve any problem. Recent dataset and model registration capabilities further enhance this functionality. If you want to customize your model, here's how! I used this to customize the model architecture specifically for pedal modeling (which also required custom C++ code to run the model in the plugin). However, the resulting Neural Amp*** Model is based on the same open source framework as a widely used standardized tool, and I was able to make this customization and quickly obtain "NAM-level" results. This was possible thanks to open source repositories.

Conclusion
I've heard the popular line that "Capture can't model knobs" or, more encouragingly, that this will be the "next frontier." It was difficult for me to navigate how to share this feature with the world, but I've had it for over a year, so I'm happy to finally demonstrate it with a free plugin. From the beginning, the purpose of sharing NAM has been to provide resources that can be used to advance the cutting edge of guitar effects and what musicians can use to create their art. I hope that by using this plugin, others will follow in this direction and continue to push the boundaries. If you are seriously interested in creating your own parametric models using NAM, please contact neuralampmodeler@gmail.com.

 

format

Download

You can download it for free.

Go to the link above, scroll to the bottom, and click the button for the appropriate OS.

Installation

When you unzip the downloaded file, the installer is included, so you can use it by simply installing it.