NAM (Neural Amp Modeler) and Proteus / Guitar ML and Amperium and ...

Discussion in 'Software News' started by ItsFine, Jul 26, 2023.

  1. ItsFine

    ItsFine Rock Star

    Joined:
    Apr 22, 2023
    Messages:
    576
    Likes Received:
    339
    NAM is well known in the guitar world.
    https://www.neuralampmodeler.com/

    But there are too :

    Proteus / Guitar ML
    https://guitarml.com/#products

    Amperium NNM
    https://www.st-rock.com/vst/

    I tried them all, and they are VERY impressive.

    Having BIG fun with Amperium "EVH 5150 III 50 RED CHANNEL" number 3 capture now.
    And NAM "Helga B 6505+ Red ch - NoBoost".

    VERY close sounding.

    I will probably stick with ONE capture i really love.
    Because there are thousands captures on NAM, witch seems the flagship.
    5 827 amps, 1 325 pedals and 1 483 irs.

    Amperium already converted some ToneX captures to their format (Total records : 1049)

    Proteus / Guitar ML : a lot less (160 captures)
    Proteus seems to sound worse than the two others. Don't know why ..

    Check them and tell me what you think :wink:
     
  2.  
  3. 洋鬼子

    洋鬼子 Producer

    Joined:
    Nov 29, 2021
    Messages:
    218
    Likes Received:
    100
    Location:
    Germany Dortmund
    So far most of my experiences with Neural Amp modellers were bad.
    I think it depends what kind of genre and style you like.
    Experimented a fuckton with AMP's and pedals my favorite ones are:

    AMPS
    -Neural DSP The nameless (Best rhythm metal AMP for lower Tunings) --->Use a multiband compressor from 100-200 hz to avoid thumping
    -Neural DSP Gojira + Bogren IR (Best Metal AMP for a more clear modern Metal Tone without too much low end but standard cab is only "okayish"
    -Neural DSP Plini (Best Clean Tone)
    -STL Tonality Andy James (Best Lead Guitar Tone)

    I heard that STL Tonality Josh Middleton is extremely nice too but never got to test it.
    Furthermore the Otto II II II II seems also really good.

    Best IR:
    Bogren IR (Anything this guy does is insanely good)
    Ownhammer

    Effects/ Pedals:
    -Valhalla Shimmer/Reverb
    -Grindstein (If you like some more Chainsaw-ish sound with a harsher feeling)
    - Soothe (Can really clean up your guitars in the 4k area)

    Most people don't understand how important IR's are to get a good guitar tone.
    In my opinion the most important part of a guitar tone.
    Did you ever made a direct comparison with an DI ?
    I saw a video of Neural amp Modellers vs Neural DSP plugins and they sounded worse to me
    Im curious how these new Neural Amp plugins compare to good Amp VST's.
     
    Last edited: Jul 27, 2023
  4. Riddim Machine

    Riddim Machine Audiosexual

    Joined:
    Jul 3, 2021
    Messages:
    632
    Likes Received:
    547
    Location:
    Jamaica
    In my experience Proteus is better emulating studio gear than NAM (NAM is mono) but NAM being superior for rigs and guitar amps than Proteus.

    Also, don't sleep on plugins using Neural AI Technology. Tone Empire was a garbage company before Neural Tech. Now their plugins sound much superior than any of those free alternatives for studio gear, also with smaller consumption than before. Don't expect light plugins (for now) using Neural Tech. I Think new Acqua uses some kind of Neural Tech on their Hyper technology also, with smaller CPU consumption than those free alternatives.
     
    Last edited: Jul 27, 2023
    • Interesting Interesting x 2
    • List
  5. Guitarmaniac64

    Guitarmaniac64 Rock Star

    Joined:
    Jun 5, 2011
    Messages:
    1,345
    Likes Received:
    317
    Well NAM is a dedicated GUITAR plugin but it can also profile studio gear but its not what it is made for at first
     
  6. StUtOpi4

    StUtOpi4 Kapellmeister

    Joined:
    Apr 9, 2016
    Messages:
    71
    Likes Received:
    60
    Ive recently got into NAM and Proteus lately and I'm addicted to modelling my hardware and other peoples hardware lol.. after modelling the VCA compressor on my Portico II Channel yesterday I ran the output model through Proteus on a drum track i'd recorded. My fucking jaw hit the floor! I have NEVER heard any plugin do what Neural modeling hardware has done to a drum track.... I literally sat in my studio chair laughing out loud every time i hit play. It was unbelievably good, as good as the hardware modelled, in fact, according to the model 0.0018 close! You will have the neigh sayers that state that "its a static" etc but it is that good that I am taking preset models for different applications, all instantly callable in Proteus as any other digital plugin preset would be in the DAW. Example, labeling it "Portico Drums - Preamp > Silk Red > AT30 > Rel100 Max 8dB GR" there are loads of world class producers (like CLA) whose hardware setting never move on certain instruments / or busses and this way of working perfectly fits that model. I wont even start to describe what I hear when I modelled the distressor!!!!!

    the other key point is that although the hardware has pre defined settings which are modelled (comp attack and release), the end sound out of the plugin model is still dynamic due to the modelling techniques, the saturation increases as you dial up the gain on the plugin and so does the amount of compression. THIS is what got me excited and will excite others of a similar ilk.

    You can apply a different model dependent on what you want to do, for example if you want to model the pre amp saturation and tube out saturation you can use a model more specifically designed for that, same for compression or cleaner type of applications

    You can tell I'm overly excited about this and I'm fairly sure that Acustica's new hyper tech is Neural too?! The way that the NAM and Proteus models capture the hardware is unprecedented and I believe its going to blow the audio world apart as more people hear and test this way of modelling / capturing for studio gear.

    We just need NAM to be be able to do stereo, both from the model input perspective and plugin output.


    Here's ChatGp's view on the million dollar question:

    does neural modelling of analog audio devices give a better result that C++ coded plugins?

    ChatGPT
    The comparison between neural modeling of analog audio devices and C++ code plugins depends on various factors and considerations. Both approaches have their strengths and weaknesses, and the choice between them ultimately depends on the specific use case, the desired outcome, and the available resources. Let's explore each approach:

    1. Neural Modeling of Analog Audio Devices:

    • Neural modeling involves using machine learning techniques, such as deep neural networks, to analyze and replicate the characteristics of analog audio devices. These models aim to capture the nonlinearities and subtle nuances that make analog equipment desirable.
    • Neural models can be effective in emulating the unique characteristics and imperfections of analog gear, such as vintage compressors, equalizers, and tape machines.
    • Neural modeling can be computationally intensive, especially during real-time processing, which may require powerful hardware for optimal performance.
    • Training neural models typically requires a large dataset of audio samples from the analog equipment being modeled, and the quality of the model heavily depends on the training data's diversity and quality.
    2. C++ Code Plugins:

    • C++ code plugins involve implementing digital signal processing (DSP) algorithms using traditional programming languages, like C++ or C. Plugin developers design algorithms that attempt to replicate the sound and behavior of analog equipment through mathematical and DSP techniques.
    • Well-designed C++ code plugins can provide high-quality emulations of analog audio devices, and they are often more efficient in terms of computational resources compared to neural models.
    • Plugin developers have more control over the design and fine-tuning of the DSP algorithms in C++ code plugins, allowing for a more tailored approach to the sound they want to achieve.
    • Unlike neural models, C++ code plugins do not require a large training dataset, but they may necessitate in-depth knowledge of digital signal processing and audio engineering principles.
    Which Is Better? The choice between neural modeling and C++ code plugins is not a matter of one being inherently better than the other. Both approaches have their merits and can yield excellent results, but they cater to different requirements and goals:

    • Neural modeling is particularly useful when attempting to replicate the subtle and complex interactions found in vintage analog equipment, where the exact modeling of nonlinearities is essential. It can provide unique character and richness that may be challenging to achieve through traditional C++ code.
    • C++ code plugins offer greater control and flexibility for developers, allowing them to fine-tune the behavior of their digital emulations. They are also more efficient in terms of computational resources, which can be critical for real-time processing and low-latency applications.
    Ultimately, the decision comes down to the specific application, the desired sound characteristics, the available development resources, and the target platform for the audio plugin. Many high-quality audio plugins are developed using C++ code, but neural modeling is becoming increasingly popular as the technology advances and more audio datasets become available for training. As both fields continue to evolve, we can expect to see further innovations in both traditional C++ code-based audio plugins and neural modeling of analog audio devices.
     
    Last edited: Jul 27, 2023
    • Like Like x 4
    • Useful Useful x 2
    • List
  7. xorome

    xorome Audiosexual

    Joined:
    Sep 28, 2021
    Messages:
    1,158
    Likes Received:
    849
    My summary after using these (they all sound great):

    - ToneX: Stereo. Light on CPU, takes way too long to start, uses too much memory, you have to feed it more highs than other modellers. Paraphrasing NNM dev: ToneX uses a lowpass to make training faster. Most polished option.

    - NAM: No stereo! Functional GUI, fast to load. Pretty heavy on CPU, nano models aren't much lighter and there are few of those anyway. Unless something's changed since February, must use 48kHz because it doesn't resample.

    - AIDA-X: No stereo! Very light on CPU, inoffensive GUI, very fast to load. Very few models. Resamples to host's rate (or says so anyway).

    - NNM: Stereo. Very light on CPU, super basic GUI. Freaks out when input is too loud. Relatively few models that aren't re-trained from ToneX/NAM. Sampling rate must match the model's sampling rate. NNM dev says: Use 44.1kHz or it'll sound wrong.

    I'm using NNM mostly because it's stereo and so light. ToneX is way too heavy for my taste. NAM is great and I'd use it if they could figure out how to make it stereo and lighter on the CPU. AIDA-X just has too few models and is mono only as well.
     
    • Like Like x 1
    • Interesting Interesting x 1
    • List
  8. Zenarcist

    Zenarcist Audiosexual

    Joined:
    Jan 1, 2012
    Messages:
    4,274
    Likes Received:
    2,737
    Location:
    Planet Earth
    I don't know how to explain it but the input volume has to align with the sampling levels, i.e. it's easy to make a good capture sound bad. Even with the software modelled amps having the right input values make a difference, e.g. for Helix Native with a Babyface Pro FS the ideal input value is +3.50 dB with zero gain on the interface preamp at -10dBV.
     
    Last edited: Jul 27, 2023
  9. inusable

    inusable Producer

    Joined:
    Jul 27, 2014
    Messages:
    111
    Likes Received:
    92
    Hello mister.
    Please could you share those "jaw dropping" models for us to try them ?
    I want to change my mind about Proteus and what you say about it is interesting.
    Thank you.
     
  10. StUtOpi4

    StUtOpi4 Kapellmeister

    Joined:
    Apr 9, 2016
    Messages:
    71
    Likes Received:
    60
    I will post a freebie later for you to try ;-)
     
  11. StUtOpi4

    StUtOpi4 Kapellmeister

    Joined:
    Apr 9, 2016
    Messages:
    71
    Likes Received:
    60
    • Like Like x 6
    • Love it! Love it! x 2
    • List
  12. ItsFine

    ItsFine Rock Star

    Joined:
    Apr 22, 2023
    Messages:
    576
    Likes Received:
    339
    Yes, there is a MASSIVE input level impact.
    Even on real hardware.

    That's why at least ONE pedal in front of amp is mandatory, just to set the input level right.
    Or at least, setting pickups height according to your amp.

    Any hardware, being amp or pedal, is created according to an average input level.
    And all amplification stages are balanced according to this input level.
    Turning the gain up or down only change ONE stage, not the balance in itself.
    Witch is fixed by design.

    Take a Fender Deluxe and input a Super Distortion in it.
    And you will hear what "too much power" means :wink:
    The opposite is true : take anaemic single coils pickups and try to get that humbucker sound into a Marshall Plexi.

    In digital world, it is even worse.
    Because levels can go to extreme.
    Especially when you chain plugins.

    PS : Thanks for all those posts mates !
    Now i understand more, especially why Proteus is not up to the game for guitar.

    I will try it against Nebula soon :wink:
     
  13. Riddim Machine

    Riddim Machine Audiosexual

    Joined:
    Jul 3, 2021
    Messages:
    632
    Likes Received:
    547
    Location:
    Jamaica

    Aw, sh*t... It seems i will need to replace the Vertigo VSC plugin on my drums. :rofl: Beautiful capture! But it would be interesting if you had a "flat" (without any compression or processing) capture of thoses hardwares. Sometimes it's better and more flexible for Proteus the "color" and natural response of the circuitry, so we can build a virtual chain with the processing with normal and colorless plugins from that. Thanks for the sharing :bow:
     
    • Interesting Interesting x 1
    • List
  14. ItsFine

    ItsFine Rock Star

    Joined:
    Apr 22, 2023
    Messages:
    576
    Likes Received:
    339
    That's why Acustica Audio disassembled some of their consoles to create sections profiling :wink:
    But not everyone is going to dismantle a multi thousand bucks console ... :rofl:
     
  15. StUtOpi4

    StUtOpi4 Kapellmeister

    Joined:
    Apr 9, 2016
    Messages:
    71
    Likes Received:
    60
    it's my pleasure to give something back. Of course your right, I am building a complete pack that will serve just that,it will take time to build up but my aim is to create "building blocks" that we can just stack, just running through the Portico II clean and adding to every track as a channel starter will transform the track on its own, then we can add saturation, Comp etc, the only issue would be running through the channel multiple times and not sure how that will stack as your going through the transformers every time, but it could turn out amazing...

    It will take a bit of work to do and get right so my first pack will be a paid pack but i'm hoping to keep costs low for those that are interested.
     
  16. StUtOpi4

    StUtOpi4 Kapellmeister

    Joined:
    Apr 9, 2016
    Messages:
    71
    Likes Received:
    60
    Im gonna put my reputation on the line here and say that this stuff beats Acustica in every department sound wise from what ive heard so far... maybe not anywhere near as refined as a complete process and usability, but for SOUND alone im finding Neural is producing better results where capturing compression, weight, saturation is concerned.
     
    • Agree Agree x 1
    • Interesting Interesting x 1
    • List
  17. ItsFine

    ItsFine Rock Star

    Joined:
    Apr 22, 2023
    Messages:
    576
    Likes Received:
    339
    I don't see any massive Proteus users push.
    For now, to be on Guitar ML website, it is mail contact ...

    Where can i get a Proteus Console ?
    Proteus Tape ?
    Any help is welcome :wink:

    PS : it is funny Proteus is BAD at guitar modeling (targeted at it) ... and best at studio gear
     
    Last edited: Jul 27, 2023
  18. Xupito

    Xupito Audiosexual

    Joined:
    Jan 21, 2012
    Messages:
    7,243
    Likes Received:
    3,997
    Location:
    Europe
    I also wonder why. I mean, probably uses a different AI model but that's still surprising.
     
  19. Riddim Machine

    Riddim Machine Audiosexual

    Joined:
    Jul 3, 2021
    Messages:
    632
    Likes Received:
    547
    Location:
    Jamaica

    On the sister site you can find some captures. And the dev (PastToFuture) is princing very cheap (15$) for what it delivers, also with some rare gear modeled, so don't hesitate to buy if you like.


    (In Italy, at this moment, there's a guy very pissed with those new Training Models:rofl:)
     
    Last edited: Jul 28, 2023
    • Agree Agree x 3
    • Funny Funny x 1
    • List
  20. Xupito

    Xupito Audiosexual

    Joined:
    Jan 21, 2012
    Messages:
    7,243
    Likes Received:
    3,997
    Location:
    Europe
    :lmao:
    He clearly needs a change of kernels
     
  21. StUtOpi4

    StUtOpi4 Kapellmeister

    Joined:
    Apr 9, 2016
    Messages:
    71
    Likes Received:
    60
    here's my new site dedicated to Analog Ai Profiles of some serious studio kit. only one set of models up right now but ill be adding more as the week goes on!

    https://analogxai.company.site/
     
Loading...