• The Gear Page Apparel & Merch Shop is Open!

    Based on member demand, The Gear Page is pleased to announce that our Apparel Merch Shop is now open. The shop’s link is in the blue Navigation bar (on the right side), “Shop,” with t-shirts, hats, neck buffs, and stickers to start. Here’s the direct link: www.thegearpageshop.com

    You’ll find exclusive high-quality apparel and merchandise; all items are ethical, sustainably produced, and we will be continuously sourcing and adding new choices. 

    We can ship internationally. All shipping is at cost.


Neural DSP Quad Cortex

thumperjack

Silver Supporting Member
Messages
712
using the knobs after a capture are supposed to do a better job of emulating what that clipping structure ‘would have’ sounded like on the source device
how long until the machine learning will walk you through a mapping process where it prompts you to slowly turn each pot min to max while it listens and reads
 

illegal gardener

Senior Member
Messages
710
Am I the only one here who doesn't care about this whole debate over the potential misuse of the words ''digital clipping'' ? And that maybe some people really seem to want to make a point on it.
You mis spelled ‘thanks gardener for getting people to google *what is guitar distortion* instead of carrying on in here’
how long until the machine learning will walk you through a mapping process where it prompts you to slowly turn each pot min to max while it listens and reads
I don’t see why it can’t be done automatically by the device, given the additional hardware needed to turn the knobs by itself - but unless the capture process can be sped up dramatically its not worth the trouble. With values from 1-10 on each of 5 knobs, at 3 minutes per capture that’s 6 hours worth of capture time for just ‘on the numbers’ combinations. When you start thinking about all the combinations between each number on each dial the number of captures and time spent gets completely absurd, even if the hardware automatically handles the naming and storing of each capture.

The answer to this problem is full circuit modeling. (For now at least)
 

thumperjack

Silver Supporting Member
Messages
712
if it was just reading frequency center and Q it could read them right quick in one sweep each and map to the capture tone knobs, you wouldn't get all the interaction between them all but at least it would let you bump the tone knobs in the capture and get basically the same freq center as the amp tone stack to use for little tweaks later. who tf knows where they get centered as generic baseline, or if NC can somehow pick up on where best to center them automatically
 

TonePilot

Silver Supporting Member
Messages
4,925
I’m losing interest in the QC. All this bickering in this thread isn’t helping to rekindle the excitement. And, the FM3 is just that good. What to do, what to do.
 

gtr37

Member
Messages
6,595
I’m losing interest in the QC. All this bickering in this thread isn’t helping to rekindle the excitement. And, the FM3 is just that good. What to do, what to do.
its like a horrible new game show where people who have no idea what they are talking about
Actually debate with engineers
Who build the products and have many years experience and knowledge about dsp and the subject .
 

doug castro

Member
Messages
514
Ok things escalated quickly.

I guess it’s better to err on the side of being overly technical than the opposite.

Let me elaborate: on previous solutions potentially not reacting too well to being pushed, my hypothesis is that given that an amplifier is a very complex system comprised of multiple filters and non-linear elements, the overall transfer function of an entire circuit can be extremely complex and even change dynamically.

This is something that a single, static non-linear element can struggle with, especially when its transfer function (curve) is predetermined.

To make matters more complex, tube amps (which are basically a power supply modulator controlled by the input signal) seem to have a time-dependent response that I believe is caused by the energy stored in the power supply’s capacitors being modulated by the pentodes and then discharged through the reactive element (speaker) not happening in zero time.
This behavior is not very well documented or understood yet, so take this as a hypothesis I’ve been working on for the past couple of years, nothing else, and thus very hard to model, and why I believe some people seem to perceive digital tube amp emulations as flat or 2D.

When thinking about how complex a system a tube amp can be from the signal processing point of view, it is not a big stretch to say that relying on a single, predetermined, static non-linearity (gain stage) for the distortion characteristics of an entire amplifier can be problematic, especially on edge cases.

One of the main challenges I have is trying to communicate the behavior of very complex systems and phenomena to a mixed audience with varying degrees of understanding of the technologies involved.
I’ll be more careful from now on.

Have a nice weekend everyone,
Doug
 
Last edited:

MrTAteMyBalls

Member
Messages
4,193
Thanks Doug! This thread went way off the rails. People care too much about semantics.

After watching the video a 2nd time I found the tones generally much better. That UI is very slick and exciting.

That first tone "mystery chode" I really don't like. This skewed my opinion of the entire video. It sounds midrangy and unpleasant. Don't like it.

Next is Rabea's capture. It sounds fantastic to me. He mentions it as a "crunch" but to me this would be about my main drive sound. Love it. Rabea's rock strat also sounds great. The ambient tones were meh. The verbs were dialed back so far you can't really hear them. Then the Freeman tone was great, as I said before, with no tweaking. I did some tracking on an HBE modded Marshall a few years ago and this is very much the sound I remember. Very cool!

Everything else was mismatch with the wrong guitar so they really didn't sound right. Not sure why he didn't just switch guitars. I am guessing he made some of those for his baritone and didn't have one available.

Anyway I just wanted to correct myself. I was a bit hard on the unit the first time.

And now back to your regularly scheduled bickering.
 

doug castro

Member
Messages
514
Thanks Doug! This thread went way off the rails. People care too much about semantics.

After watching the video a 2nd time I found the tones generally much better. That UI is very slick and exciting.

That first tone "mystery chode" I really don't like. This skewed my opinion of the entire video. It sounds midrangy and unpleasant. Don't like it.

Next is Rabea's capture. It sounds fantastic to me. He mentions it as a "crunch" but to me this would be about my main drive sound. Love it. Rabea's rock strat also sounds great. The ambient tones were meh. The verbs were dialed back so far you can't really hear them. Then the Freeman tone was great, as I said before, with no tweaking. I did some tracking on an HBE modded Marshall a few years ago and this is very much the sound I remember. Very cool!

Everything else was mismatch with the wrong guitar so they really didn't sound right. Not sure why he didn't just switch guitars. I am guessing he made some of those for his baritone and didn't have one available.

Anyway I just wanted to correct myself. I was a bit hard on the unit the first time.

And now back to your regularly scheduled bickering.
Hey man, thank you, I appreciate that.
I think things will be clearer to everyone who hasn’t had a chance to try it first hand once more videos start going live.
 

doug castro

Member
Messages
514
You mis spelled ‘thanks gardener for getting people to google *what is guitar distortion* instead of carrying on in here’

I don’t see why it can’t be done automatically by the device, given the additional hardware needed to turn the knobs by itself - but unless the capture process can be sped up dramatically its not worth the trouble. With values from 1-10 on each of 5 knobs, at 3 minutes per capture that’s 6 hours worth of capture time for just ‘on the numbers’ combinations. When you start thinking about all the combinations between each number on each dial the number of captures and time spent gets completely absurd, even if the hardware automatically handles the naming and storing of each capture.

The answer to this problem is full circuit modeling. (For now at least)
That’s a very astute and lucid comment actually.

A few other tough technical problems aside, the possible number of combinations alone can be incredibly problematic (and you do to measure the interactions between knobs, which is something tube amplifiers are known for).

Say, for argument’s sake, that you need to capture a knob in 6 different settings to measure its behavior (you can then simply interpolate between these settings and get a fairly accurate recreation of its continuous behavior).
The number of possible combinations will scale exponentially the more controls you have, if “n” is the number of controls, and you need 6 positions, you’ll need 6 to the “nth” power combinations.

So for one knob it’s obviously 6 positions, not too bad.
2 knobs 36, a pain in the rear but still manageable.

A full amp with 7 knobs (gain, bass, mid, treble, master, presence and depth)?

6 to the 7th power, which is 279,936 possible combinations.
 
Last edited:

illegal gardener

Senior Member
Messages
710
That’s a very astute and lucid comment actually.

A few other tough technical problems aside, the possible number of combinations alone can be incredibly problematic (and you do to measure the interactions between knobs, which is something tube amplifiers are known for).

Say, for argument’s sake, that you need to capture a knob in 6 different settings to measure its behavior (you can then simply interpolate between these settings and get a fairly accurate recreation of its continuous behavior).
The number of possible combinations will scale exponentially the more controls you have, if “n” is the number of controls, and you need 6 positions, you’ll need 6 to the “nth” power combinations.

So for one knob it’s obviously 6 positions, not too bad.
2 knobs 36, a pain in the rear but still manageable.

A full amp with 7 knobs (gain, bass, mid, treble, master, presence and depth)?

6 to the 7th power, which is 279,936 possible combinations.
Thanks, and exactly- this is totally doable- it’s just time prohibitive to the point it would probably only be useful as a dev tool - maybe to true-up a model. I’m assuming your approach to full circuit modeling already involves a similar machine learning process at the component level anyway so this probably wouldn’t contribute much to the end results of your models.

BUT, the idea that with a cortex and a dongle of leads to attach to the knobs of the source device -that this process could become streamlined to the point end users could create their own full circuit models on the device by capturing all these combinations is fascinating. ‘Folding for the tone’ instead of ‘folding for the cure’
 
Last edited:




Top