NDSP Quad Cortex

[laxu]

You can read about NeuralDSP's approach here: https://arxiv.org/html/2403.08559v1

They don't use just X fixed values for each knob but a randomized sample. While this might mean that the model is not accurate at every single possible permutation, in practice it's probably in that "nobody cares" territory where you can't really notice it.

 

[Lysander]

Well, that explains the number of new amp models in their recent firmware updates. Maybe they built TINA without doing the math in advance.

What's funny to me is how NDSP annonuced this TINA thing almost three months ago, only to later note that CorOS 3.1.0 is including no new amps

 

[mbenigni]

What's funny to me is how NDSP annonuced this TINA thing almost three months ago, only to later note that CorOS 3.1.0 is including no new amps

It was also really weird that they made the TINA announcement at the same time as the big PCOM (finally) announcement.

"We built a massive, insanely expensive robot that can turn physical knobs so that we're better prepared to convert our software plugins to run on a different processor." Hmm.

 

[Lysander]

It was also really weird that they made the TINA announcement at the same time as the big PCOM (finally) announcement.

Yeah, they even sort of tied the two together, which was odd.

 

[mbenigni]

You can read about NeuralDSP's approach here: https://arxiv.org/html/2403.08559v1

They don't use just X fixed values for each knob but a randomized sample. While this might mean that the model is not accurate at every single possible permutation, in practice it's probably in that "nobody cares" territory where you can't really notice it.

Are we talking about random "seed" values, with subsequent values chosen according to extent to which deltas have an impact? Otherwise, I can't see any value in randomizing versus using an even distribution (however coarse) across the full range. Random distributions might be better, and they might be worse... at random.

(Confession: I did not read the white paper.)

(EDIT: OK, now I've skimmed it. It reads to me like they're intentionally not using random settings, but rather using an even distribution - with whatever level of granularity is "affordable" - transversed via a "traveling salesperson" algorithm. Which is something I'd never heard of until this minute. This still fails to answer @Orvillain's point above, which is that even extremely coarse distributions of controls will yield extremely long capture times.)

 

[spawnofthesith]

NQCD for my band mate

 

[Achilles]

BASIC NASTY.

Kinda tells you everything you need to know about that preset!

 

[Jarick]

@2dor legend has it that you've got a solution to make sure you get the right amount of gain from a Neural capture, so that you don't lose gain in the capture when compared to the guitar going into the amp???

That's my one big issue with QC, the gain staging is weird for captures. You always seem to lose gain. Best I did was to dial in the sound running the pedal or amp in the effects loop, then switch it over to capture and get it as is. That seemed pretty close. Anything else was off.

That issue seems like it would be straightforward for NDSP to solve, but here we are years later...

 

[duzie]

It was also really weird that they made the TINA announcement at the same time as the big PCOM (finally) announcement.

"We built a massive, insanely expensive robot that can turn physical knobs so that we're better prepared to convert our software plugins to run on a different processor." Hmm.

The robot

 

[Achilles]

View attachment 29228
The robot

Flashback.....

 

[duzie]

Fixed it

 

[Warmart]

Damn, I am getting gassy for a QC.

If they get all the Fortins converted I just may jump on one.

 

[eggpl@nt]

Damn, I am getting gassy for a QC.

If they get all the Fortins converted I just may jump on one.

I’ve been enjoying the QC more the second time around. It’s still far from perfect and I have a general distaste for NDSP as a company, but fun tones is fun tones and more toys is more toys.

 

[Orvillain]

NQCD for my band mate

View attachment 29219

That is pretty cool.

I've been envisioning a board with the QC and my Meris LVX and MercuryX, plus maybe some kind of midi controllable drive or boost pedal, all in 4-cable-method with my Dual Rectifier and a midi switchy box, wah pedal, volume, and an expression pedal.

 

[Achilles]

I've been envisioning a board with the QC and my Meris LVX and MercuryX,

I got excellent results putting Strymon stuff in the loop(s) of the QC!

 

[Mongillo19]

That is pretty cool.

I've been envisioning a board with the QC and my Meris LVX and MercuryX, plus maybe some kind of midi controllable drive or boost pedal, all in 4-cable-method with my Dual Rectifier and a midi switchy box, wah pedal, volume, and an expression pedal.

I struggled with the H9 in the loop and getting levels right. But you are smarter than me with this stuff

 

[fiftyshadesofjay]

I also gave up on using it in 4cm with my Recto as it sounded terrible so the odds may be stacked against him here

I struggled with the H9 in the loop and getting levels right. But you are smarter than me with this stuff

 

[yeky83]

It takes about 3 minutes to capture an amp using QC. Assuming their internal tech takes the same amount of time, and let's assume that the amplifier has these knobs:
Gain
Bass
Middle
Treble
Depth
Resonance
Master

And assume you're capturing 10 positions of each knob. That would be 10,000,000 permutations. It would take approximately 57 years of continuous profiling to capture all 10 million permutations, assuming it takes 3 minutes per permutation.

If you assume 3 positions of each knob, it would take approximately 4.56 days of continuous profiling to capture all 2,187 permutations, assuming it takes 3 minutes per permutation.

So I'd think interpolation would almost certainly be necessary, even with robotic automation of knob positions in sync with a scripted recording+capturing system.

Skimmed through their paper and it says they modeled the DC-30 with five knobs using below method:
“Each example in the dataset consists of a pair of one second long input-target audio segments, sampled at 48 kHz. The control positions are kept constant over each segment, and stored together with the audio, as described in section 2.1. Input audio sequences were drawn randomly from a collection of guitar and bass recordings. The dataset totals around 4.5 hours of paired audio, randomly split into 15000 training and 1000 validation examples.”

Their language with pairs is confusing, seems like around 6 or 7 positions per knob was used.

 

[Lysander]

Their language with pairs is confusing, seems like around 6 or 7 positions per knob was used.

Note that this describes only data collection - which is, comparatively, the easy part. No details on how they actually model this stuff, other than LSTMs being used.

 

[Orvillain]

There's definitely funky stuff going on with levels with the QC. It is actually quite tricky to wrap my head around. There's quite a lot of things to account for.

XLR outputs are +9.5dBu.
TRS 3+4 outputs are +15.5dBu if you use balanced cables. +9.5dBu if you use unbalanced cables. This is the first source of oddity.
The Send 1+2 outputs are +9.5dBu.

I've connected my Meris MercuryX to the send 1+2 and return 1+2. I've got the Meris setup for true bypass operation currently. I get a level increase when I turn the effects loop block on inside of the Quad Cortex. Which isn't really desirable. I can't say I've had this problem with the Helix or Axe3. I can double check, but I'm pretty sure I can just put a pedal into the loops on those units and get unity gain.

So there seems to be a bit of a juggling act with levels, and tbh, I don't even know where to start with optimizing the signal-path in order to get the right level hitting the front of outboard effects, the right level at the returns of the QC, the right level going into the PA without being too low, and the right level coming out of outputs 3+4 to send to on-stage power amplifiers.

With the Axe3 for instance, I can take a feed from output 3... turn the output 3 level knob on the front all the way up, and I know it is the same level hitting the front of a guitar amp as what it would receive if I just plugged my guitar directly into the amp.

The same does not seem to be the case for the Quad Cortex. I cannot trust that send 1 is the same level as what the amp would see if I was plugged directly in. And debugging this stuff is a mare.

Again to repeat, in 4-cable-method the QC is noticeably noisier than the the Helix and Axe3.

You can get a good sound. You can whack a noise-gate after the effects loop block to cut down any extra noise that the QC adds. But you don't need to do that with the other two devices.

It is very odd that the documentation doesn't really point any of this out, and doesn't give you any tools to improve the signal to noise ratio and to ensure unity-gain throughout the system. The Axe3 truly does.

My expectation is that you connect the thing up to a valve amp in 4-cable-method, leave all of the built in trims at 0dB, and there should be no level drops or boosts, and there should be no significant additional noise added.