Project LYDIA, a concept prototype by Roland Future Design Lab, uses Neutone’s machine learning technology to convert any input into a “tone.” But this is not just about “modeling”. Everything can be handled by something else. Inputs can be anything from beatboxing to field recordings. It may be the opposite of what genAI sounds like, due to its focus on sampling and real-world tinkering. This is technology that requires you to go out and play.
Roland has a decades-long history here. First introduced in 1981, AMDEK’s “Analog Music Digital Electronic Kits” DIY pedals were a low-cost way for modding enthusiasts to acquire pedals. So we hope this signals a return to AMDEK tradition.
The 80s meets 2020s here, as the Project LYDIA prototype has a Raspberry Pi 5 board inside. (Audio is currently connected externally via USB, but we plan to add hardware to make these jacks work and make it self-contained.)
Transferring sounds in such an anything-goes style is much more interesting than a typical demo. Many early machine learning demos touted using AI to model vintage analog circuits. The problem is that there are already countless hardware and software tools to do it, and humans are very good at sitting next to the board and tweaking the code with subtle nuances.
But this is something your average pedal can’t do. This is most evident when you watch their demo video.
And this seems to give a better vision of what this content is like. Instead of using screen time to write prompts, like when you’re trying to argue with an airline’s automated support bot, we can go out and do what we want with sound: listen, record, and play. So of course it’s great that Roland put this in a classic metal box.
This is not a replacement for human ability. Because, like processes like granular synthesis and convolution, it’s a digital process that opens up sounds that can’t be created any other way. (That being said, humans are humans, and we find ways to imitate things once we hear them. So don’t say never. It certainly ventures into territory where BOSS pedals have never gone before.)
The enabling technology comes from the Neutone Mopho, and you don’t need to use this pedal’s concept to make it happen. Their plugins are available and quite attractive. But I think you’ll agree with me in this video that I perform best when I’m doing the least common things.
It’s also important to note that this type of training can be performed using very small data sets and can be worked locally without using the cloud. This is similar to the difference between a V12 car and a bicycle, which guzzle leaded gasoline. Both are means of transportation, but one can be more harmful than the other.
Here’s a closer look at the state of the already well-received hardware prototype at Brighton’s now legendary Audio Developers Conference, where Project LYDIA was announced yesterday.
Due to the nature of data science, I would really like to see an experimental and open approach to this. AI in its current state, including commercial applications, exists because of these experiments. So, if you’re itching to try something like this yourself; keep it up. In short, that was the very spirit that created guitar pedals in the first place, including a healthy level of competition.
But I think there’s more to say on this subject, especially since I’ve been speaking with Roland’s Paul McCabe, who heads up the Roland Future Design Lab.
P.S. Enjoy the AMDEK pedals here. (I mean, this pedal is what I would come up with. Well, yes. I would like this more than the TR-1000.)
Things have worked out well for Sega when it comes to incorporating the idea of intelligent listening into their products. I mean, at all I didn’t Although I work for them, I have a Dreamcast and a copy of Seaman, so I thought it would be interesting to look back at this ad.
more:
https://neutone.ai/fx
Introducing Project Lydia [Roland]
