Roland Future Design Lab, the advanced technologies research group of Roland Corporation, and AI developer Neutone have introduced Project LYDIA, an audio processing platform that turns a Raspberry Pi 5 into a compact neural sampling audio processor.
Powered by Neutone’s Morpho technology, Project LYDIA, named after DIY and AI, uses “neural sampling” to learn the tonal qualities of any sound and apply it to another in real time, with the sonic results ranging from highly musical to exotic and surprising.
The current Project LYDIA mockup houses a Raspberry Pi 5 computer running Neutone’s special version of Morpho alongside Roland’s latest ideas for a simple, tactile control panel. Audio I/O is managed via a USB-connected Roland Rubix interface, with future iterations aiming to fully integrate I/O for a completely self-contained device.
To learn more, see the the Roland site.
Okay, I want one to make crazy stuff with my voice
Looks very similar to IRCAM’s acids-ircam RAVE project. Which also runs on pi as well as pd and max and you can dl and play with now for free.
https://github.com/acids-ircam
So, basically a VST in a box?
I suppose ALL digital effect pedals are “VSTs” in a box, if “VST” is your shorthand for DSP on a computer.
This is unique because it presents new processing capabilities in a pedal format (with an R-Pi 5!) FIRST. I would love a plugin version of this, but because it works in realtime, a hardware fx unit makes sense.
the funny thing is they dont really say what the processing actually is. ^^
(dont watch the man behind the curtain) ^^
I watched some of their videos and all they said was “vocoder” & “neural sampling”,
I guess they thought ppl understand what a vocoder does and dont understand what convolution can do?
So they dont say “dynamic convolution” …
but it strongly sounds like that to me.
the box is just a playback device,
you need to create “the thing” on your separate computer and then give it to the box.
creating “the thing” seems pretty involved.
feed me hours of audio material …
not just 5 or 10 minutes
this wont be for everybody.
“Project LYDIA, named after DIY and AI”.
Then where did the L come from then?
the L is from Lab. It’s written right on top fofthe pedal
No, sorry. I think you’re wrong.
It definitely from Lettuce.
Lettuce. DIY. AI. 202 (because everything Roland has to have a 0 in the middle).
Lettuce.
Those sounds are crazy cool. This something truly new, and though the sounds are sort of outrageous, they are completely useable, IMHO.
Glad to see this with Roland, as they have the muscle to make it pretty robust.
This use of AI as a “random in context” for sounds, melodies and patches its ok (like Scaler but more intuitive, guided and automatic). Hoever that “make me a song like thig artist and that artist” is whack.
Seems like you are referring to AI’s use in generative composition/audio/song-making processes. That’s not what this is.
You imply that Scaler uses AI, but I don’t think that is true. Unless something changed recently (?)
I think this is an effect processor that uses some new kinds of processes to alter the input signal. This isn’t convolution, but has some things in common with that, I suppose.
I wonder if they use AI to help generate the DSP code that is then run on the Raspberry Pi, or if the R-Pi is actually doing the neural sampling in the box?
hm, it all stands or falls with whether you have enough interesting models to play with in the end or not.
So it’s basically a preset machine.
feeding the network with 1 ½ hours of samples to create your own “model” is way to involved for the average Joe from nextdoor, imho.
it smells like lets sell stuff via Roland cloud. :/
there is no display to show you the name of the model loaded.
it could be anything.
this is just a “blackbox”, thats not helpful.
This is an early prototype, not their final release version. I can’t imagine they’d release it without a display.
A Raspberry Pi 5 has “Dual 4Kp60 HDMI® display output with HDR support”. So you can connect it to anything with a HDMI port. Is your TV not “display” enough?
I am certainly not interested in looking at my TV to get the name of my “musical playground currently in use”
Roland does the hardware box for this
they asked for input
include a display
is my 2 cents 😉
Roland is making new experiments, thats very interesting. 🙂
Their old wine in new skins isnt very interesting to me.
im not sure what “neural sampling” … to learn the tonal qualities of any sound and apply it to another in real time,
is supposed to mean?
it sounds like marketing speak for “some kind of dynamic convolution” ???
if you know enlighten me.
I thought the very same.
Reinvention of the Hartmann Neuron only pared down to four knobs, and so more affordable and available. But, it has been done before.
I was enthousiastic at first but i also thought this is finally some marketing answer to the lightyears more forward thinking Korg.
But i heard a lot of ‘bla bla’ and the demo’s sounded ‘fart fart’.
(Not completely unlike korgs mechanical synth demoos 😉 but Korg is still lightyears more inventive.
Both mentioned products have a potential for gas, but sorry this sounds duff.