ML Sound Lab - Amped Volcano [Parametric NAM amp modelling]

[2dor]

NAM's making it to the top. The latest Amped amp sim plugin (based on the Soldano SLO-100 and the Soldano Astro) uses parametric NAM technology to model the amplifiers.

Amped Volcano - ML Sound Lab

Amped Volcano is based on a classic American high gain amp, as well as its modern counterpart. This is our first plugin to run on VornAI which uses advanced AI and deep learning to recreate the sound of these sought-after guitar amps with impressive accuracy.

ml-sound-lab.com

Mikko's been very clear in their use of NAM & a shout out to Steve (NAM creator):

 

[Mongillo19]

Very cool! It does sound a bit stringy compared to what I think of on a SLO30 (which I know isn't the same amp)

I'll give it a look

 

[2dor]

I just took it for a quick spin & it feels very good under the fingers.

Wonder what other amps they will release.

 

[nicolasrivera]

The one that i loved was the ASTRO Blue. It sounds and feels just like captures NAM.

 

[Dom Mcsweeney]

Sounds great, and I'm excited to see where this goes in the future, but my only gripe is that currently the settings of parameters from one amp channel duplicate when switching to a different amp channel - which is more than a tad infuriating.
Those settings should not be universal, but individual per channel.

I've messaged Mikko about that, hopefully he'll get that sorted out.

 

[GTR37]

Be interesting to see how this compares the NDSP SLO
Plug-in

 

[TheTrueZoltan!]

I downloaded the demo yesterday evening and really like the SLO, but haven‘t spent much time with the Astro model yet. As @Mongillo19 already stated, the SLO sounds slightly stringy, which is something I usually don‘t like, but somehow with this plugin it doesn‘t really bother me. I've always found the modeling of the power amp saturation to be a weak point in ML Sound Lab plug-ins - the master control never reacted the way I would actually expect it to. Apparently ML saw it the same way and has now simply done away with the master control, although this is probably due to the new hybrid modeling engine. The amp is presumably captured with a fixed master volume setting. Overall I really like the plugin and will most likely hit the purchase button today. Oh, and I really enjoy the virtual pedalboard of the ML plugins. It offers exactly what I usually need, not more, not less. And the effects sound great and are easy to dial in.

 

[[Nathan]]

Just gave the demo a spin, feels like a significant upgrade to the ML plugins.
The SLO is a tough buy because so many existing plugins/suites exist, but still a good option out there. I use the NDSP one a lot and have little to no interest in the newer Mercuriall one or this ML sound one for multiple reasons (which is basically if it aint broke...).

Mikko said this was a sort of half step into what they plan to do so im excited to see what they've got cooking next. I really hope they update their player because mono only in 2025 is kind of lame and little things like pressing the pedalboard and then it not disappearing off screen unless you specifically press the pedalboard button again etc etc, the interface just needs a facelift and some creature comforts imo.

Either way this is a massive step forward for them, good to see them adapting and using newer/better tech

 

[MirrorProfiles]

Hang on, it doesn’t have a master volume? I’d have thought that’s something that NAM does very well and relatively easy parametrically. There’s a bazillion decent SLO emulations as is, IMO it would have been worth getting the MV behaviours locked in. It makes a big difference in an SLO

 

[2dor]

Hang on, it doesn’t have a master volume? I’d have thought that’s something that NAM does very well and relatively easy parametrically. There’s a bazillion decent SLO emulations as is, IMO it would have been worth getting the MV behaviours locked in. It makes a big difference in an SLO

Absolutely agree - that amp has a lot to give across a great deal of its MV range.

That said, the way Mikko describes it, they wanted to push it out the door as a PoC of sorts.. guess there's more to come.

 

[MirrorProfiles]

Absolutely agree - that amp has a lot to give across a great deal of its MV range.

That said, the way Mikko describes it, they wanted to push it out the door as a PoC of sorts.. guess there's more to come.

Interesting. I'd imagine for a fully parametric model, the best way is to separate the preamp and poweramp and have them as separate models.

If this is an older SLO, the FX loop is before the tone stack so you can't easily separate them easily to do captures, unless you mod it (but you'd kind of want to leave the original FX loop in place and add a transparent one after the tone stack).

Proof of concept is cool, but then it's also a product to buy. Seems more like a demo/light version that they'd probably want to go over once they have this stuff figured out. Cool they're working towards something and trying to do things in a more in depth way to their previous models. That poweramp behaviour is a must really for a "full" model IMO.

 

[BenIfin]

Just gave the demo a spin, feels like a significant upgrade to the ML plugins.
The SLO is a tough buy because so many existing plugins/suites exist, but still a good option out there. I use the NDSP one a lot and have little to no interest in the newer Mercuriall one or this ML sound one for multiple reasons (which is basically if it aint broke...).

Mikko said this was a sort of half step into what they plan to do so im excited to see what they've got cooking next. I really hope they update their player because mono only in 2025 is kind of lame and little things like pressing the pedalboard and then it not disappearing off screen unless you specifically press the pedalboard button again etc etc, the interface just needs a facelift and some creature comforts imo.

Either way this is a massive step forward for them, good to see them adapting and using newer/better tech

Interesting. I'd imagine for a fully parametric model, the best way is to separate the preamp and poweramp and have them as separate models.

If this is an older SLO, the FX loop is before the tone stack so you can't easily separate them easily to do captures, unless you mod it (but you'd kind of want to leave the original FX loop in place and add a transparent one after the tone stack).

Proof of concept is cool, but then it's also a product to buy. Seems more like a demo/light version that they'd probably want to go over once they have this stuff figured out. Cool they're working towards something and trying to do things in a more in depth way to their previous models. That poweramp behaviour is a must really for a "full" model IMO.

Id be interested in your thoughts on this video Mikko posted.



The clean / crunch / edge of breakup tones in ^this^ video .... are just a horrendous mess of top end fizz and sandpaper crackle - either he has no idea how to dial in this Amp (?) -or- there are insane levels of very audible aliasing (?) -or- maybe both (?)

 

[Orvillain]

there are insane levels of very audible aliasing (?) -or- maybe both (?)

This is one of the downsides of neural network based amps.

Neural networks learn a mapping from input to output audio without understanding the underlying signal processing or physics.

Aliasing happens when a signal contains frequencies higher than half the sample rate (the Nyquist limit), and those high frequencies get "folded" back into the audible range as false, harsh-sounding tones.

Unlike traditional amp models that explicitly oversample or apply anti-aliasing filters when generating those harmonics, ML models usually don’t know they’re creating frequencies that will alias. They just reproduce whatever they saw in the training data. If the model was trained at 44.1kHz and you ask it to process a signal at 44.1kHz, it may produce aliased harmonics with no protection. The network doesn't inherently understand or filter anything unless it's been built or trained to do so.

If you increase the DAW sample rate (say from 44.1kHz to 96kHz), you give the model more “headroom” before aliasing occurs. That means the same high-frequency harmonics that would alias at 44.1kHz might now be preserved cleanly at 96kHz. This can reduce aliasing, even if the model itself isn’t oversampling—it just happens because the Nyquist limit is now higher.

But if the ML model was trained at a fixed sample rate (say 44.1kHz) and you feed it audio at 96kHz without adapting or retraining it, the model may behave unpredictably. Some models are sensitive to sample rate, and the learned behavior might not generalize well. You could get incorrect tone, broken dynamics, or even worse aliasing, depending on how the model processes time and frequency.

And yes, that video sounds pretty shit to me.

 

[riffy]

The master volume in an SLO, makes a huge difference to me. I find my SLO doesn’t really come alive until it hits about four on the master volume.

 

[Dom Mcsweeney]

The clean / crunch / edge of breakup tones in ^this^ video .... are just a horrendous mess of top end fizz and sandpaper crackle - either he has no idea how to dial in this Amp (?) -or- there are insane levels of very audible aliasing (?) -or- maybe both (?)

Seems to be using a fair few guitars - including with p90s and single coils, which will sound brighter.

 

[laxu]

Seems to be using a fair few guitars - including with p90s and single coils, which will sound brighter.

It still sounds to me like some sort of aliasing noise rather than e.g single coil noise.

 

[Orvillain]

Kind of hilarious how it sounds no better than Amplitube 2 from like, 2006.

 

[[Nathan]]

Id be interested in your thoughts on this video Mikko posted.



The clean / crunch / edge of breakup tones in ^this^ video .... are just a horrendous mess of top end fizz and sandpaper crackle - either he has no idea how to dial in this Amp (?) -or- there are insane levels of very audible aliasing (?) -or- maybe both (?)

I dunno about aliasing but that jazzmaster clip is probably the biggest offender of what you're talking about. Who knows how much processing / IR swapping and all that went into this, very hard to say. I checked the demo for 15 minutes and it didnt seem egregiously different to the NDSP one. I get that stringy sort of sound if the gain is too low on their input knob, if you guess around (which I dont like) you can find a point where it mostly goes away, thats the point I started to match it to NDSP (I still couldnt match it cause I have to guess their input gain and guess their MV).

But without being too much of a nerd, it was doing a good job. Just hard to do a side by side comparison.