I made a plugin called Gear Capture. Anyone want to give it a try?

[BenIfin]

...... we have NAM A1 and soon'ish A2, Tonex V1 and V2, QC V1 and V2, Kemper V1 and someday (?) V2 .... and Proxy is going to drop soon'ish which may or may not also be NAM-native ready, and Fractal have already said their next Gen unit(s) will be NAM compatible.

Plus numerous 2nd tier Hotone, Nux etc.... with their own stuff too.

Its going to have be to provably and audibly orders of magnitude better than all of the above just to even get to the changing room before the starting line.

But I admire their hutzpah ... I think

 

[BenIfin]

Regarding compression, I know everyone uses it differently, but I still think capturing a single state makes sense. I've heard that some mixing legends, like CLA who is famous for rock, rely heavily on fixed compression settings. Also, if you use a lot of modeled compressors and compare them to the actual hardware, you'll see the tone is usually way off. Our captures actually sound better.

As for amp gear, I'm a little lost. We can get test results that trade blows with full-size NAM using half the CPU, but I don't know if that's enough to get people interested.
Since I'm a guitar player myself, I've noticed issues with other products that are hard to explain with simple specs. Things like unnatural playing feel or weird high-frequency noises. I've done a ton of research and optimization on this stuff. Unfortunately, I haven't figured out an easy way to explain it yet, and there's no simple metric to measure it. That's why I really recommend you guys try our plugin for yourselves. Of course, I'm not sure if this is what you all care about.
So, other than CPU usage and some really subtle differences, I can't seem to find any other direction for optimizing amp modeling right now.

Video ?

 

[FractalAudio]

I mean 0dBFS equals 12.2dBU. I checked dozens of different guitars and their RMS levels usually fall between -30 and -18.

dBFS mean dB re. Full-Scale. dBu means dB unloaded. You cannot convert between dBFS and dBu. dBu is an absolute scale with 0 dBu being 0.775V RMS.

dBFS is relative to the maximum voltage the A/D converter can convert without clipping. This voltage depends on the reference voltage. Some converters have a reference voltage of 2.5V, some have 3.3V, some have 5V, etc.

RMS means Root Mean Square. It is typically a voltage. You square the signal, take the average and then take the square root. It is impossible to have a negative RMS value.

Looks like you guys might be doing some good work but if you want to be taken seriously you should at least understand these fundamental aspects of audio engineering.

 

[PineappleKing]

I dont understand, how is this different or Better than NAM?

We have a different design philosophy and tech path, which gives our gear capture a few advantages:

- It uses less CPU for the same level of accuracy. I have to be clear that this is partly because we use more data and a more complex optimization process. This does make training and capturing take longer, but that is exactly what we intended.
- You can use oversampling to control aliasing.
- It does a better job capturing the behavior of compression over time.

Coming up next:
- Single knob parameter modeling.
- Train once and it works natively on all common sample rates.

 

[PineappleKing]

Video ?

What kind of videos are you looking for? I actually have a compression comparison test in the videos I've already uploaded.

 

[HotRats]

There are outliers

Yes there are and that's why a blank statement is incorrect.

Keeping the gain at 0 is not the recipe for the best experience possible, it's a good starting point.

When accurancy is the goal you can't gain stage things randomly hoping for the best.

 

[Sascha Franck]

So, other than CPU usage and some really subtle differences, I can't seem to find any other direction for optimizing amp modeling right now.

For a start, a plugin with a nice featurset to load your captures would possibly interest some folks. It certainly would trigger my interest.

 

[MirrorProfiles]

Yes there are and that's why a blank statement is incorrect.

It’s a generalisation.

Keeping the gain at 0 is not the recipe for the best experience possible, it's a good starting point.

It depends, but with the gear being described in the generalisation it probably IS the best experience possible.

When accurancy is the goal you can't gain stage things randomly hoping for the best.

It’s not random at all. It’s taking into account the fact that lots of gear is using 8-13dBu=0dBFS for their instrument inputs. If you can’t improve SNR further, and you have a consistent baseline for ALL instruments, what exactly are you optimising for?

 

[PineappleKing]

dBFS mean dB re. Full-Scale. dBu means dB unloaded. You cannot convert between dBFS and dBu. dBu is an absolute scale with 0 dBu being 0.775V RMS.

dBFS is relative to the maximum voltage the A/D converter can convert without clipping. This voltage depends on the reference voltage. Some converters have a reference voltage of 2.5V, some have 3.3V, some have 5V, etc.

RMS means Root Mean Square. It is typically a voltage. You square the signal, take the average and then take the square root. It is impossible to have a negative RMS value.

Looks like you guys might be doing some good work but if you want to be taken seriously you should at least understand these fundamental aspects of audio engineering.

Okay, let me explain the situation a bit more.

I used an oscilloscope and a signal generator to confirm I was getting an 8.925V (peak-to-peak) sine wave at 1000Hz. Then, I adjusted the gain on my audio interface until it hit 0dB in my DAW. After that, I recorded clips with dozens of different guitars and used a script to calculate their RMS values in dB, which led me to that conclusion.

Looking back, I realize this wasn't very rigorous, and we definitely should have done a better job.

 

[MirrorProfiles]

Only saying this constructively, not to be overly critical or anything. This was an example from the website that read extremely AI generated (they’re consecutive sentences too). If it isn’t AI, I’d still recommend changing it to something that has less hallmarks of AI generated text. It’s also very waffly - it’s not really saying much.

We didn't just copy it; we dug deep to understand the "logic" behind its sound.

LittleFuzzE isn't just a cold piece of software. It's the soul of that golden age of rock, reborn inside your computer.

 

[PineappleKing]

Only saying this constructively, not to be overly critical or anything. This was an example from the website that read extremely AI generated (they’re consecutive sentences too). If it isn’t AI, I’d still recommend changing it to something that has less hallmarks of AI generated text. It’s also very waffly - it’s not really saying much.

Got it. I'll definitely make those changes. It's kind of funny though because I actually rewrote this so many times. I guess I'm just not cut out for marketing.

 

[PineappleKing]

dBFS mean dB re. Full-Scale. dBu means dB unloaded. You cannot convert between dBFS and dBu. dBu is an absolute scale with 0 dBu being 0.775V RMS.

dBFS is relative to the maximum voltage the A/D converter can convert without clipping. This voltage depends on the reference voltage. Some converters have a reference voltage of 2.5V, some have 3.3V, some have 5V, etc.

RMS means Root Mean Square. It is typically a voltage. You square the signal, take the average and then take the square root. It is impossible to have a negative RMS value.

Looks like you guys might be doing some good work but if you want to be taken seriously you should at least understand these fundamental aspects of audio engineering.

Replying again. I'm super excited about your response because I'm a total Fractal fan.

 

[HotRats]

It’s a generalisation.


It depends, but with the gear being described in the generalisation it probably IS the best experience possible.


It’s not random at all. It’s taking into account the fact that lots of gear is using 8-13dBu=0dBFS for their instrument inputs. If you can’t improve SNR further, and you have a consistent baseline for ALL instruments, what exactly are you optimising for?

If you leave the preamp gain at 0 you will likley need to correct the input at the plug-in, somentimes by a lot, depending on you hardware.
Differences in audio interfaces specs (or modellers used as audio interfeces) can be easlily of more than 6dB.

I'm sure we agree that to get the best experience you have to make the specs of the interface the plug-in match.

I agree it's a generalization, an unecessary one. Not a big deal since it's not a product manual just a post on a forum.

 

[MirrorProfiles]

If you leave the preamp gain at 0 you will likley need to correct the input at the plug-in, somentimes by a lot, depending on you hardware.
Differences in audio interfaces specs (or modellers used as audio interfeces) can be easlily of more than 6dB.

Absolutely. And if you have the same reference point for all guitars, you can just save your input gain in the plugin to whatever value you need for each one and they're always ready to go. If you're always adjusting your input level by different amounts, then you need to change it each time in each plugin. And if you're jumping between sessions, getting back to the same input levels is a bit of a chore. So if you aren't improving your SNR in any way, I think its much much better to use a fixed level.

I'm sure we agree that to get the best experience you have to make the specs of the interface the plug-in match.

I agree it's a generalization, an unecessary one. Not a big deal since it's not a product manual just a post on a forum.

Agree on both. I just think there really needs to be a very good reason not to record with gain at 0, rather than the other way around. Sometimes there is a good reason but these days I think these days it's largely a solved problem. I say this as someone who doesn't use input gain at 0 (as Im using an external DI with its own level controls).

 

[DrewJD82]

Welcome to the forum @PineappleKing!

These guys can be tough critics but you’ll get some great feedback to work with, I don’t think I’ve come across a group that’s more serious about the spec side of things than the guys here! Good luck!

 

[nicolasrivera]

We have a different design philosophy and tech path, which gives our gear capture a few advantages:

- It uses less CPU for the same level of accuracy. I have to be clear that this is partly because we use more data and a more complex optimization process. This does make training and capturing take longer, but that is exactly what we intended.
- You can use oversampling to control aliasing.
- It does a better job capturing the behavior of compression over time.

Coming up next:
- Single knob parameter modeling.
- Train once and it works natively on all common sample rates.

Ok, thank you. Its more clear now. Good luck.

 

[PineappleKing]

Hey everyone. Here is a look at the tasks we are focusing on right now. Please let us know if you have any better ideas or if there is a feature you really want to see sooner.

To-do list (sorted by priority):

1. Experimenting with datasets for better generalization.
2. Cleaning up the official website.
3. Adding AAX support.
4. Making a high gain amp demo.
5. Polishing the automatic null test tool for release.
6. Single training run with native support for any sample rate.
7. Parameterized modeling for single button.
8. Automatic and manual gain staging.
9. Side chain compression.
10. Stereo support for mono compression captures.
11. Mid-side support.
12. Mod capture.

 

[2dor]

Hey everyone. Here is a look at the tasks we are focusing on right now. Please let us know if you have any better ideas or if there is a feature you really want to see sooner.

To-do list (sorted by priority):

1. Experimenting with datasets for better generalization.
2. Cleaning up the official website.
3. Adding AAX support.
4. Making a high gain amp demo.
5. Polishing the automatic null test tool for release.
6. Single training run with native support for any sample rate.
7. Parameterized modeling for single button.
8. Automatic and manual gain staging.
9. Side chain compression.
10. Stereo support for mono compression captures.
11. Mid-side support.
12. Mod capture.

Sounds great.

IMHO the gain staging / calibration should be on the top 3 given how so much of the overall user experience leans on it.
Right after it, I'd say parametrized single-knob training.

 

[Orvillain]

@PineappleKing you should consider adding local CUDA training. I have an NVIDIA 4090, and I don't want to train in the cloud.

 

[PineappleKing]

@PineappleKing you should consider adding local CUDA training. I have an NVIDIA 4090, and I don't want to train in the cloud.

Oh, I've written a ton of custom operators and heavily optimized them for the latest GPUs, so porting everything to a general version would be a huge project. Also, the VRAM required for training right now is massive. When the model size and parameters are set high, even the memory on a 4090 or 5090 won't be enough.

I'll definitely consider it, but right now I have more important things on my plate.