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

[PineappleKing]

View attachment 57950

My bad, maybe I didn't explain that clearly. When I say correct gain staging, I mean 0dBFS equals 12.2dBU. This makes sure that on most common audio interfaces, you can just set the gain to 0 and get the best experience. Based on that standard, I checked dozens of different guitars and their RMS levels usually fall between -30 and -18. So aside from me not explaining it well the first time, I think the idea holds up.

Unless maybe the gear I tested wasn't diverse enough? I would really love to see some more professional data on this.

 

[[Nathan]]

I’d be interested in how the amp captures sound vs nam vs tonex v2 vs the real amp. I think if you can show people some kind of tangible improvement it would be a good selling point

 

[PineappleKing]

I also don't want to be a buzz kill... but I cannot tell what the plugin does either, and your description of oversampling versus "ultra" oversampling leaves a lot to be desired.

And if you are going to say things like this:


Then you really need to back it up with some evidence. The page is called "Our Technology" but it doesn't describe the technology in anything but the most high level and basic way. Not very useful.

I'm also not joining a Discord to get a plugin.

You are right. I totally realize now that I didn't explain things well enough before.

I did touch on ultra oversampling a little bit in my latest video. But I am planning to make a dedicated video that goes into way more detail about it soon. I am honestly really proud of this tech. As far as I know, DeepSoundLab is the only one doing this right now.

For the other stuff, I feel like the current market vibe is partly to blame. Let me vent for a second here. There is just so much fake marketing out there. You see buzzwords like component-level modeling everywhere, but very few products actually deliver on that promise. That is just not how I operate. I only promote what I have actually built. I buy other audio gear too, and I really hate false advertising.

 

[PineappleKing]

I gave it a go. It wasn't an enjoyable experience, honestly.

It took 2 hours between uploading the audio file to getting a model trained.
That's pretty much a waste of a Saturday morning.

The model sounded off, so I decided to look at it a bit more detailed:

Null test was approximately -16 LUFS, which is pretty bad compared to what this is up against, not that it really matters if there's something else happening such as heavy oversampling.

My main concern here isn't even that, but everything else about the venture feels off (lack of solid technical information / snake oil vibes / chatgpt-esque descriptions and responses in the forums / a lack of transparency of who the real person(s) are behind everything / the discord funnel / the email funnel ... the bragging without real information to back it up).

It's a bit of a nightmare to me.

I was approached on Facebook by the same person(s) before this new capturing direction a few months back, when it was just about the component modelling, begging me to allow them to post in a facebook group I run (which I declined for the same reasons and gripes).

It's all a mystery to me how people lap this up (or perhaps they were born yesterday?).

---------------

Emails showing evidence of the time it took:

View attachment 57957



Some pretty pictures that probably don't really matter, comparing a NAM model here to this new tech:

LUFS: -16.4
EQ and Phase look very different.

View attachment 57956

Meh.

I don't get why you'd think I sound like I'm using ChatGPT. I typed every single word myself. Sure, since English isn't my first language, I used a translation tool a bit to help out. But looking at what I wrote, I think it's all very solid and full of real substance. I really don't believe ChatGPT could come up with something that logical and packed with that much specific professional knowledge.

About your training session this time, I am truly sorry you had a bad experience. There are actually a lot of reasons for this. Let's walk through the whole process, from the training all the way to the final null test.

First off, right now we only support "dist" and "comp" models. I'm not sure exactly what gear you used to capture the data, but I analyzed it carefully. The characteristics look a lot like a linear time-invariant system—what we usually call an IR or digital EQ kind of thing. Because of the specific tech route we chose, we can't support that type of gear very well just yet. Honestly, if it really is a simple linear time-invariant system, I'm guessing a lot of products out there wouldn't support it well either. Designing the optimization process is super complex. Trying to use a model with non-linearities to fit a linear system usually results in pretty bad gradients. Of course, this is just a guess. I would love to know what specific gear you used because we've trained dozens of devices so far and have never seen this happen before.

Second, the long training process is actually part of Gear Capture's design philosophy. Unlike other products, I prefer using a ton of data and longer training times to get better accuracy without using up more CPU. We actually have a hyperparameter grid search option in the works that isn't finished yet, and that's going to take even more time.

Finally, after the training is done, we calibrate the phase to always be positive. We do this to make the "amount" and "mix" features work better. Unfortunately, your target gear flips the output phase, so the null test didn't run correctly. I've noticed that everyone really loves doing null tests, so I actually built an automatic null test tool. I'm going to polish it up and release it soon. That should save everyone a huge amount of time.

 

[HotRats]

This makes sure that on most common audio interfaces, you can just set the gain to 0 and get the best experience

If you set the interface gain at 0 it works but it will hardly give you the best results/experience.
The best way to operate is to set the preamp gain as high as possible (avoding AD clpping and preamp distrotion) and then, after the AD conversion, trim the level to match the plug-in specs.

 

[MirrorProfiles]

My bad, maybe I didn't explain that clearly. When I say correct gain staging, I mean 0dBFS equals 12.2dBU. This makes sure that on most common audio interfaces, you can just set the gain to 0 and get the best experience. Based on that standard, I checked dozens of different guitars and their RMS levels usually fall between -30 and -18. So aside from me not explaining it well the first time, I think the idea holds up.

Unless maybe the gear I tested wasn't diverse enough? I would really love to see some more professional data on this.

Claiming that any kind of recording levels are a professional standard is a bit strange, let alone by using RMS, and especially for using RMS on a guitar DI. It doesn’t really give any helpful information (is the guitar part a Brian Eno song or a Metallica song?).

SNR and headroom are important but there’s a lot of factors at play that’ll determine where your optimal levels are, and they’ll be case dependent. So you can’t just give a value that is optimal like some kind of cheat code.

For digital gear, dBFS is probably the most important scale to reference. RMS isn’t going to be useful really when trying to set an optimal level for an instrument.

And maybe most importantly, for modelling analog gear you would want to capture things relative to a fixed analog reference point. That way the plugin can be developed and used with a consistent response no matter who is using it.

 

[PineappleKing]

I’d be interested in how the amp captures sound vs nam vs tonex v2 vs the real amp. I think if you can show people some kind of tangible improvement it would be a good selling point

From a standpoint of copyright and respect, I don't think it's right for us to go out and do this kind of thing ourselves. That said, we'd be more than happy to let someone else handle it.

Technically speaking, I actually think everyone is doing a great job. With these kinds of products, especially the ones using real deep learning, you can usually get awesome results if you're willing to put in the training time and ramp up the CPU usage to handle more parameters and data. Different products just have different design philosophies, so comparing them directly often isn't very fair.

I'm a big fan of using an accuracy-to-error ratio as a standard, rather than just looking at the error rate alone. Otherwise, some dishonest folks could totally cheat by using massive parameters and huge datasets for their own models, while testing others on tiny parameters and small datasets.

I do have an idea about this, though. It would be amazing if a third-party company or an influencer could host a modeling competition, kind of like Kaggle. Let the product designers enter the contest themselves. Then we could compare their error-to-performance ratios and do a big subjective listening test to rank everyone.

 

[MirrorProfiles]

Just a few other thoughts.

what % of ToneX/NAM/Tonehub/Tonocracy/Kemper/QC users have an issue where accuracy is holding them back? Kemper for sure has some instances where it struggles badly, but I’d imagine most of those users have moved onto something else. Anyone using a Kemper in 2026 clearly doesn’t care about having something more accurate with any kind of urgency.

Same goes for aliasing and oversampling? Aside from a very vocal but minute number of forum dwellers, I think most users not only don’t have a problem with aliasing, but they wouldn’t even reliably be able to identify it. I’m not saying it can’t be improved (or shouldn’t be) but I also don’t think it’s a bottleneck for the vast majority.

As for improvements, I think I’ve written lots on forum posts already and I’m not sure how well they’d work on this anyway. IMO I’d prefer this to focus on either compression or guitar amps and make distinct products for each.

As for compressors….. how many models would I need to make to cover all the settings I like to use? What about to have the same kind of flexibility as something like rcomp or any stock compressor plugin? If there isn’t the full range of controls available, and if it takes a lot of time and effort to make them, what exactly is the incentive to use this tech to begin with?

For the guitar amps….. you need a very very good reason to make people jump into a proprietary/closed platform. Aside from people who care a lot about aliasing or oversampling, why would anyone jump into this? You’d really need to demonstrate what the existing platforms pitfalls are and why that matters.

 

[[Nathan]]

The thought of using a new platform just reminds me of how much time I’ve spent with tonex, tonex v2 and nam.

Going through that thing again is not a fun task.

As a side note I really hope stadium proxy is pretty faultless, I’d hate to have to work its quirks out and second guess if the results are better/worse than the competition and if it’s worth doing tons of captures all over again.

Pretty much applies to anything new coming along, all going to have the same sort of “Is this worth it” from a capturers pov.

 

[MirrorProfiles]

If you set the interface gain at 0 it works but it will hardly give you the best results/experience.
The best way to operate is to set the preamp gain as high as possible (avoding AD clpping and preamp distrotion) and then, after the AD conversion, trim the level to match the plug-in specs.

At 12.2dBu, how much gain do you expect to be able to boost? and how much is that going to improve SNR on modern converters? They’ve made it optimal by design.

Not saying there aren’t edge cases, but OP did say for most users on interfaces with 12.2dBu. It’s a sensible level and often used for a reason.

 

[2dor]

The thought of using a new platform just reminds me of how much time I’ve spent with tonex, tonex v2 and nam.

Going through that thing again is not a fun task.

As a side note I really hope stadium proxy is pretty faultless, I’d hate to have to work its quirks out and second guess if the results are better/worse than the competition and if it’s worth doing tons of captures all over again.

Pretty much applies to anything new coming along, all going to have the same sort of “Is this worth it” from a capturers pov.

Yeah, I mean my hands are already full with NAM lol

Anything else coming along expecting wide adoption must prove itself to be pretty damn good - better than NAM (much better probably) - for most to jump onboard until the platform matures.

 

[Sascha Franck]

Quick question, as I wanted to give this a try: Is this really limited to 48kHz? I mean, for capturing, that'd be kinda fine with me, but for playback, I'd prefer to stick with 44.1 for the time being as pretty much every project I ever made is using it.

Anyhow:

what % of ToneX/NAM/Tonehub/Tonocracy/Kemper/QC users have an issue where accuracy is holding them back?

Every bit this, a very valid question.
I think that especially NAM and Tonex are pretty close to the diminishing-return-realm already when it comes to accuracy (take that with a grain of salt as most of you should already know that I'm not too much into accuracy).

Add to this that at least right now, NAM is offering you some already pretty mature options to utilize things (well ok, IMO all the player plugins urgently need some database/file management, but that should be a piece of cake in case someone will finally take it seriously), the combination of freely available plugins and the super easy to deal with capturing with the help of Tone3000 really sets a bar that would need to be taken.

Heck, while not super accurate, my 119 bucks Valeton GP-50 loads NAM files (I know, it wouldn't pass any accuracy test - but it's still pretty damn impressive), my (by now 3) Tonex Ones, whille certainly having some issues, are the most excellent "utilitarian" units I possibly ever bought and instantly became the sound backbone of my live setups.

These are the things any new player would have to compete with.

As for compressors….. how many models would I need to make to cover all the settings I like to use?

Another valid question. I'm not much into compressors but typically use the Logic ones, but even I have a plethora of HQ compressors in my arsenal - all of which I've got for free or free-ish. And there's new free ones on offer pretty much any day.
Add to this that with an amp capture, I typically can get a lot out a single capture by just the gain controls and whatever post-amp tonestacks (it only gets better with dual tone stacks such as in Genome). I don't think you can get that much out of a compressor capture because the only thing you'd be able to tweak in a static capture would be input gain (hence threshold) - but with compression, even as someone not much into them, I very often use at least the attack and ratio parameters, something that simply can't be part of any single capture process.
So why would I want a statically captured compressor, unless it was some super rare vintage thing doind something absolutely special?

Having said all that, I think the next big step up would either be parametric capturing or, somewhat related (and of much more interest for myself), some kind of engine that would allow me to morph/interpolate between pretty much any two (or more) captures.
Yet another step forward (but I don't have the slightest idea whether and how that would be possible in the not too distant future) would be being able to capture, say, modulations *and* being able to at least somewhat tweak them in the resulting capture.

At least these would be what I'd consider to be interesting to a mere end user like me.

 

[HotRats]

At 12.2dBu, how much gain do you expect to be able to boost? and how much is that going to improve SNR on modern converters? They’ve made it optimal by design.

Not saying there aren’t edge cases, but OP did say for most users on interfaces with 12.2dBu. It’s a sensible level and often used for a reason.

I'm just saying that the blank statment is wrong.
IF your signal needs 0 dB boost than so be it, IF NOT than turn up the gain knob because is there for that porpouse.

 

[MirrorProfiles]

I'm just saying that the blank statment is wrong.
IF your signal needs 0 dB boost than so be it, IF NOT than turn up the gain knob because is there for that porpouse.

Don't want to derail, but his point is valid. Most interfaces are designed for optimal guitar input levels with gain at 0 these days, and thats exactly the point he is making. There are outliers (like Behringturd which is either way too hot for hum buckers, or needs a pad+gain) but thats why it's a generalisation.

A vast % of users are using something from Focusrite/UA/Audient/RME/SSL/Arturia. Increasing gain is unlikely to improve SNR, and thats in the situations where it wouldn't lead to clipping. Im mostly a passive humbucker user and I can clip most of these inputs with gain at 0 (exception being SSL).

 

[Orvillain]

Increasing your interfaces gain also raises noise-floor. So quite why people keep giving the advice that you should record as hot as possible without clipping, is anyone's guess. It just isn't necessary and you're not getting any "better" an experience by doing so.

Arguably you're getting a worse experience, because if you record with multiple guitars and multiple pickups, you're essentially discarding the level differences between them if you're always aiming to be as hot as possible.

 

[Sascha Franck]

Fwiw, totally idiotic OT side comment: Interface trim at 0, dialing down your volume pot and being a Logic user = a match made in hell.
Wondering if they finally fixed that "< -35dB peaking files not properly accepted" bug in Logic 12 (due to be released on the 28th).

 

[[Nathan]]

if you record with multiple guitars and multiple pickups, you're essentially discarding the level differences between them if you're always aiming to be as hot as possible.

That old chestnut. It’s almost like a single coil Strat is going to be a wee bit quieter than fishman fluence active pickups that are so hot they self clip on chugs (dunno if they actually clip or limit but they brick wall in some fashion)

 

[PineappleKing]

Quick question, as I wanted to give this a try: Is this really limited to 48kHz? I mean, for capturing, that'd be kinda fine with me, but for playback, I'd prefer to stick with 44.1 for the time being as pretty much every project I ever made is using it.

Anyhow:



Every bit this, a very valid question.
I think that especially NAM and Tonex are pretty close to the diminishing-return-realm already when it comes to accuracy (take that with a grain of salt as most of you should already know that I'm not too much into accuracy).

Add to this that at least right now, NAM is offering you some already pretty mature options to utilize things (well ok, IMO all the player plugins urgently need some database/file management, but that should be a piece of cake in case someone will finally take it seriously), the combination of freely available plugins and the super easy to deal with capturing with the help of Tone3000 really sets a bar that would need to be taken.

Heck, while not super accurate, my 119 bucks Valeton GP-50 loads NAM files (I know, it wouldn't pass any accuracy test - but it's still pretty damn impressive), my (by now 3) Tonex Ones, whille certainly having some issues, are the most excellent "utilitarian" units I possibly ever bought and instantly became the sound backbone of my live setups.

These are the things any new player would have to compete with.



Another valid question. I'm not much into compressors but typically use the Logic ones, but even I have a plethora of HQ compressors in my arsenal - all of which I've got for free or free-ish. And there's new free ones on offer pretty much any day.
Add to this that with an amp capture, I typically can get a lot out a single capture by just the gain controls and whatever post-amp tonestacks (it only gets better with dual tone stacks such as in Genome). I don't think you can get that much out of a compressor capture because the only thing you'd be able to tweak in a static capture would be input gain (hence threshold) - but with compression, even as someone not much into them, I very often use at least the attack and ratio parameters, something that simply can't be part of any single capture process.
So why would I want a statically captured compressor, unless it was some super rare vintage thing doind something absolutely special?

Having said all that, I think the next big step up would either be parametric capturing or, somewhat related (and of much more interest for myself), some kind of engine that would allow me to morph/interpolate between pretty much any two (or more) captures.
Yet another step forward (but I don't have the slightest idea whether and how that would be possible in the not too distant future) would be being able to capture, say, modulations *and* being able to at least somewhat tweak them in the resulting capture.

At least these would be what I'd consider to be interesting to a mere end user like me.

Right now it only supports 48kHz, but we are working on a way to make it automatically handle all sample rates with just one training session.
We are also working on parameterized capturing for single knobs, but that might take a while.
A few users asked about capturing mod effects recently, so we added that to our to-do list.
It is mostly just me working on this project. I have a friend who helps out in his spare time, but we cannot move as fast as the big companies. Thanks for being patient with us.

 

[nicolasrivera]

View attachment 57949





Hey everyone,

We just finished up a new plugin called Gear Capture, and it’s now in closed beta. You can probably guess what it does from the name. We’ve noticed that the current options out there still feel pretty rough around the edges, with a lot of common requests and issues left unresolved. So, we decided to take a crack at it and offer our own solution.







Just so you know, this is an invite for a private test. It is not the official launch yet, so everything is free. Honestly, most features will likely stay free even after we go live.

Now Supported

- Ultra Oversampling. We didn't just implement oversampling, we took it a step further than the standard implementations out there. You can check out the tech page at https://deepsoundlab.com/tech to see exactly how our Ultra Oversampling works.
- Lower CPU usage, higher precision. It runs more efficiently while giving you even better audio quality.
- Precise system response. I think we are all realizing that a null test isn't the whole picture. Matching the impulse response and harmonics is just as important. Gear Capture delivers a more accurate system response. It brings a sense of realism to those subtle details and honestly just feels better to play.
- Flexible settings. You get to decide how to balance capture quality against CPU load based on what your project needs.
- Compression capture support. The precision here is leading the industry right now.
- Accurate GR meter display. When you are using compression capture, the gain reduction reading is spot on.
- Low Latency. Getting captured data to line up perfectly with the original audio is usually tough, but thanks to our unique phase alignment tech, Gear Capture has lower latency than anything else out there.
- Amount and Mix Knobs. Because we’ve got such a precise handle on the phase of the capture, the Amount and Mix features on Gear Capture work exactly the way they should.

Coming Soon (We have the tech ready, we just need a little time to finish the engineering)

- Full sample rate support. You will be able to train once and run it at any sample rate. Just to be clear, this isn't simple resampling back to the capture rate. It is genuine, native support for any sample rate.
- Full single knob capture. This allows you to sample the entire range of a single knob yourself, rather than just capturing a static snapshot.
- Side chain compression.
- Stereo support for mono compression captures.
- Mid-side support.

Like the title says, this isn’t an official launch yet since we are still in the testing phase. If you want to chat more about it or give Gear Capture a spin, come hang out with us on our Discord channel.

Join the DeepSoundLab Discord Server!

Check out the DeepSoundLab community on Discord - hang out with 9 other members and enjoy free voice and text chat.

discord.com

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

 

[PineappleKing]

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.