Recently I became aware of a rather new technique for modelling guitar amps, overdrive pedals and other non-lineair equipment. It’s called black-box modelling. Instead of modelling all the components of a piece of hardware (virtual analog design), modelling the op-amps, the filters and so on (which we call white-box modelling – the ‘we know it all approach’), the black-box method is not trying to find out what exactly is going on inside the box but only tries to find out what comes out of the box. It uses convolution techniques, but in a non-lineair way, to measure how the hardware changes the input into output.
Recently research has been done by Aalto University and Neural DSP Technologies. The name Neural says it all, since this technique is based on using neural network models. But as far as I know Neural DSP Technology haven’t used this technique in a plugin yet.
“Deep neural networks for guitar distortion modeling has been tested before,” explains Professor Vesa Välimäki of the work, “but this is the first time where blind-test listeners couldn’t tell the difference between a recording and a fake distorted guitar sound! This is akin to when the computer first learned to play chess.”
https://www.hackster.io/news/smart-black-box-neural-networks-recreate-classic-guitar-amp-sounds-in-real-time-0c5d2156607f
You can find the PDF of the full-text of this research here:
https://www.mdpi.com/2076-3417/10/3/766
As an additional bonus the researchers found out that the neural network (black-box) models used way less CPU power to run compared to their white-box virtual analog modelled counterparts.
All the models included in the listening tests were shown to be capable of running in real-time on a consumer-grade desktop computer, with some of the models only requiring a relatively small amount of the computer’s processing power.
By the way, a mix between white-box modelling and black-box modelling can also be useful. This is called grey-box modelling.
If I am correct Kemper Amps is also using black-box techniques for profiling amps. And the profiling technique Overloud uses in their TH-U plugin is based on the same technique. However in the research paper I found no references to these profiling techniques. Many guitar players find the Kemper profiles extremely realistic. In this paper though the blind-test listeners couldn’t tell the difference between the real amp and the black-box version.
A well done A/B listening test is the best way to achieve objectiveness. And we need it because our brains are very subjective. This is often the problem with guitar players who don’t like amp simulation and/or impulse responses. Because they know they are not using the real thing, their brains doesn’t allow them to enjoy the sound of it.
Guitar players are often extremely conservative.
Anyway, this sounds very promising to me. In the near future it will be reasonably easy to profile an amp using these techniques. As if you’re making a photocopy of your equipment 🙂
UPDATE: check out the super interesting article by Keith Bloemer about GuitarML, his open source, community driven project that uses machine learning to model real amps and pedals.