Get with the programmers
Can plugins be ‘intelligent’? The ex-Eventide engineer and Newfangled Audio founder certainly thinks so…
Newfangled Audio products are “combining DSP technology with advances in the field of machine learning.” For the uninitiated, please explain…
DG “When you brush away the marketing buzzwords, this is really about new techniques for solving problems with more than one right answer. The question becomes: how do you choose the best answer out of a number of good answers? In Elevate, we break the signal up into a number of frequency bands and need to set the level for each one so that the sum of these bands doesn’t go above the ceiling. So what should the levels be? There are a lot of right answers, but we want the one that gives you the loudest output while satisfying other criteria that we know make a signal sound good. When you translate those ideas into math, it becomes a machine learning problem, and we can use machine learning techniques to get the best answer.
“The other nice thing about the machine learning approach is that these criteria become the adaptive parameters, so the user can still have control over the machine learning algorithm.”
Elevate, Equivocate and Punctuate use linear-phase filters to divide the signal into 26 ear-sympathetic frequency bands (known as the Critical Bands). How did this concept evolve, and how did this approach affect the development, coding and/or testing process?
DG “Circuit modelling has been a big topic for plugin developers recently, but why are we spending so much effort modelling analogue circuits when we could be directly modelling the human ear? It’s undeniable that some analogue circuits sound great, but it’s also true that those classic processors were built with analogue circuits because that’s what the creators had access to, not because it was always the best tool for the job. I started looking into modelling the ear because I’m looking for areas where we can do better than what the best analog circuits have given us.
“The science behind the Mel Scale and the 26 Critical Bands goes back to the 1930s at Bell Lab, and the auditory models that have come from it are used in speech detection, audio coding and other audio technologies. I would encourage readers to go read the Wikipedia article on Critical Bands – it’s short, and you might learn a lot about musical perception.”
“This is about solving problems with more than one right answer”
What other ‘intelligent’ technologies are you using? DG “The intelligence in these tools really boils down to: 1. Model the human ear to create a natural representation of the audio you want to process, and 2. Use machine learning to (attempt to) model the human brain to make good decisions about how best to process audio. There are a couple of other neat DSP tricks here and there, but when I say ‘intelligence’ in the marketing, I really do mean that, so that it can pick the best answer all the time.” What’s next for Newfangled Audio? DG “I’m trying to figure that out at the moment. I’ve got a bunch of ideas and projects that I’ve started, but I want to make sure that whatever is released next will be useful to people while also pushing the technology.”