Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Show HN: SmartGuitarAmp – Guitar plugin made with deep learning

github.com

Post reply on HN
Print view View original

151–160 of 163 posts

Re: Show HN: SmartGuitarAmp – Guitar plugin made with deep learning

#151

Post by klodolph »

irscott wrote: post #137
klodolph wrote: post #117

Earlier quoted context omitted.

Crank your AC15 and play it in the basement, it will destroy your hearing pretty fast without protection. I used to take a homemade 5W amp outside and hear it echo in the woods. Fun times. Quick math—the Vox AC15 has speakers with around 100 dB SPL@1m,1W sensitivity, so if you’re cranked then it’s +24dB, and if you’re 2m away then that’s -12dB, and that’s 112 dB SPL, somewhere around “car horn” or “chainsaw” on the l…

Why would you need to crank it at all? That amp is gonna sound fantastic at half volume. For grit they make excellent overdrives these days.

It takes all kinds. If you’re running your tube amp below breakup, you might as well just use a solid state amp. It’s only under overdrive conditions that you get compression. Outside overdrive conditions, there is not really any clue that you are using a tube amp. Mostly linear is mostly linear.

It could be that you just like the volume. The speakers in the AC15 are fantastic. Or it could be you don’t think of it as gritty, but I do.

Re: Show HN: SmartGuitarAmp – Guitar plugin made with deep learning

#152

Post by fuzzfactor »

klodolph wrote: post #101
qppo wrote: post #91

Earlier quoted context omitted.

Yes you have more control, it also sounds worse. They have advantages in live reproduction, yet you still see tubes mic'd up in the studio. And for a small band even if you have a FoH mix, it's going to sound better out of a 40W combo than through whatever PA your vocalist dragged along. But that's just my opinion. Digital models do a good job in a vacuum (no pun intended). They do a very poor job of modeling how a g…

> …lower impedance inputs of a pure tube stage… I don’t know what your idea of “lower impedance” but a pure tube stage has an input impedance that measures close to open. Off the range of my multimeter, I bet. You just see the Miller capacitance and with the grid-plate somewhere around 1.6pF and maybe 20dB gain, it’s not a lot. The input impedance of the guitar amp is usually dominated by the 1 MΩ resistor to ground,…

Yes the input impedance of a good tube is much higher than the typical 1 Megohm resistor referencing it to ground.

That's a million ohms so not much of your guitar signal is wasted passing through it to ground.

The input impedance of the tube reacts dynamically according to the signal coming in, and the often dynamic behavior of whatever is plugged directly into it.

When that's just your guitar without any pedals or anything, that can be the most dynamic from a touch-sensitive point of view.

Re: Show HN: SmartGuitarAmp – Guitar plugin made with deep learning

#153

Post by analog31 »

TheOtherHobbes wrote: post #146
IndySun wrote: post #131

Earlier quoted context omitted.

"...Take it as a fact that you’ll commonly have higher standards than your audience. This applies to software, actors, musicians, whatever..." Can you explain this a bit more? I do not see this is commonly true, it is the exception and not the rule in many cases of 'someone and their audience'. It might be true that that 'someone' believes it, however misguided.

Any form of art or engineering requires a huge amount of background knowledge and/or practice experience, and artists and engineers should have a better grasp of the technicalities than the audience. But audiences tend not to care about specifics. They care if you can live up to expectations and provide the experience they expect. So if a performance is lacklustre, or a piece of software is stupidly designed from the…

Indeed, and I'm conscious of this as an amateur with a pro-league instrument. ;-) I spent many years playing a student grade instrument, had used it for countless performances, and nobody ever complained about my sound.

But a nicer instrument is a pleasure to have, and provides me with the assurance that anything wrong with my sound is strictly a matter of improving my own abilities. So it's a challenge to live up to the quality of my instrument.

Re: Show HN: SmartGuitarAmp – Guitar plugin made with deep learning

#154

Post by oriolid »

qmmmur wrote: post #149
oriolid wrote: post #132

Earlier quoted context omitted.

Convolution is linear, tube amps have many levels of nonlinear behavior.

Whether or not it is nonlinear is another question. I'm saying that amp emulation using convolution is cheaper computationally, usually a lot more approachable and way more configurable. Ultimately people use this to make sound, so I'm commenting on the sound qualities that convolution affords which to me are good enough considering you've already gone digital to model something analog anyway.

The thing is, convolution based guitar amp simulation is not going to sound like guitar amp, because convolution can't model the nonlinearity. You might get otherwise interesting sounds, but not what is usually considered the electric guitar sound.

Re: Show HN: SmartGuitarAmp – Guitar plugin made with deep learning

#155

Post by sickcodebruh »

albutr wrote: post #148
sickcodebruh wrote: post #145

Earlier quoted context omitted.

Here’s my unscientific breakdown of how I view the Nolly, Cali, and Nameless plugins. Nolly: most versatile thanks to its multiple amps, far and away my favorite for most things. I recently used it to mix an entire album of very polished black metal, replacing the real fuzzy mic signals captured in the studio. The band loved it. Extremely detailed. Got some gorgeous cleans out of it, too. Cali: this one reminds me a…

awesome, thanks for the breakdown! Any experience with Plini? It seems to be the closest to Nolly, and from a cursory look it seems people generally like it better. I suppose I'll have to get the trials for both to compare I had never heard of GGD's IR loader, seems awesome! Definitely on the list of things to buy now Feel free to drop a link to the album you made with the nolly plugin, I'd love to take a listen

I haven’t used the Plini but only because I hit Nolly last and I’m thrilled with it. I’m pretty sure the Nolly rhythm amp is modeled after a 5150 or a 5150II, so it’s very familiar for me since I mostly work with extreme metal.

I seriously can’t say enough good things about Zilla Cabs. Just make sure to disable the cabs in the amp plugin or it’ll sound weird!

The album I did with the Nolly plugin isn’t something I can share yet, unfortunately.

Re: Show HN: SmartGuitarAmp – Guitar plugin made with deep learning

#156

Post by valdiorn »

This will get lost in the pile, but I've done some tests with a dumb-as-a-brick neuralnet, 5 layers of 128 nodes, reLU, and the thing shows INCREDIBLE quality. I build VST plugins as a hobby, and have done several amp and guitar effects plugins. I also build and design tube amps.

I was very pleasantly surprised at the quality I was getting, and it was simple enough to run in real-time using pytorch (well I exported the model to onnx and used a c++ lib to run it). I was on my way to making a little startup company around the idea when covid hit and priorities radically changed

Re: Show HN: SmartGuitarAmp – Guitar plugin made with deep learning

#157

Post by dmacvicar »

whiddershins wrote: post #133
dmacvicar wrote: post #77

Earlier quoted context omitted.

Many prominent artists are touring with modelling amps for years now. https://www.fractalaudio.com/artists/ There are many advantages. Less maintenance, lower stage volumes, better for in-ear-monitoring use, sound consistency, less weight to carry, and the ability to explore hundred of amps sounds with a single device eg. check https://wiki.fractalaudio.com/wiki/index.php?title=Amplifier...

Linking to an endorsement page doesn’t tell the whole story. I would be pretty surprised if Metallica were using the technology you are describing. They have an extremely consistent setup that has worked for decades and an effectively infinite budget. That’s just one concrete example. I think a lot of the comments in this thread are missing the fact that if a player thinks a thing has irreproducable properties, they…

Metallica IS using this in every single concert they have played after they experimented with them in the Antartica show. AxeFX II, but I saw AxeFX III in a recent photo.

https://www.youtube.com/watch?v=P6k2ghtbhoU 3:09-7:32.

Megadeth is also using AxeFX. https://www.youtube.com/watch?v=O7vzOjzn0Mw

And from there you can extrapolate how common they are.

Re: Show HN: SmartGuitarAmp – Guitar plugin made with deep learning

#158

Post by albutr »

sickcodebruh wrote: post #155
albutr wrote: post #148

Earlier quoted context omitted.

awesome, thanks for the breakdown! Any experience with Plini? It seems to be the closest to Nolly, and from a cursory look it seems people generally like it better. I suppose I'll have to get the trials for both to compare I had never heard of GGD's IR loader, seems awesome! Definitely on the list of things to buy now Feel free to drop a link to the album you made with the nolly plugin, I'd love to take a listen

I haven’t used the Plini but only because I hit Nolly last and I’m thrilled with it. I’m pretty sure the Nolly rhythm amp is modeled after a 5150 or a 5150II, so it’s very familiar for me since I mostly work with extreme metal. I seriously can’t say enough good things about Zilla Cabs. Just make sure to disable the cabs in the amp plugin or it’ll sound weird! The album I did with the Nolly plugin isn’t something I ca…

I just spent a couple of hours playing with the nolly trial and it's awesome, lots of fun. Definitely getting some 5150 vibes from several of the presets, I believe it

Re: Show HN: SmartGuitarAmp – Guitar plugin made with deep learning

#159

Post by copperx »

valdiorn wrote: post #156

This will get lost in the pile, but I've done some tests with a dumb-as-a-brick neuralnet, 5 layers of 128 nodes, reLU, and the thing shows INCREDIBLE quality. I build VST plugins as a hobby, and have done several amp and guitar effects plugins. I also build and design tube amps. I was very pleasantly surprised at the quality I was getting, and it was simple enough to run in real-time using pytorch (well I exported t…

Would the neural net be useful for modeling other hardware? Or is it configured specifically for guitar amps? If so, would it be able to model something like a hardware compressor?

Re: Show HN: SmartGuitarAmp – Guitar plugin made with deep learning

#160

Post by dmacvicar »

TheOtherHobbes wrote: post #143
hashkb wrote: post #85

Earlier quoted context omitted.

I just returned a Helix because of lackluster... everything. I'll try modelers again in a few years. I think of tube amps a lot like I think about Vim.

The Helix is honestly pretty awful, so I'm not surprised you returned it. I spent some time auditioning modellers a few years ago, and my reaction to the Helix was "LOL! Seriously??" The Kemper is far more credible, with the AxeFX system and and some of the better amp sim plugins you can buy from (say) UAD following behind.

How did you audition them? what did you use in addition to the modeler?
Post reply on HN