AI Amp Modeling Next Gen: A Frankenstein Scenario

[agedhorse]

is this why it is legal for an amp modeler to copy a popular amplifier’s tone with impunity? Or isn’t it?

Just note that they can't use legal trademarks of other companies without license.

 

[J Wilson]

Why not?

Against the backdrop of the celebrations currently along the lines of '50 Years of HipHop', you have several generations of people who've only ever heard assembled music from samples with little or no live playing added in.

So it's a dog chasing it's tail: Automatic music creation for people that only know that kind of music.

Now I'm going back to my daybed and wait for AARP to call me back. BTW, get off my lawn.

 

[ardgedee]

Just note that they can't use legal trademarks of other companies without license.

This has led to one thing that endlessly irritates me, though: Euphemisms for brand names in modelers.

My Fender amp has models for "Rumble", "'59 Bassman", and "SWR Redhead". Cool so far. And then it has models for "KGB 800", "Monster" and "Rockin' Peg". uh, okay. One's reasonably obvious (although not particularly specific), the others I still haven't figured out.

Unless you're already thoroughly familiar with the original equipment there's no way you'll know what some settings are supposed to be. And it's way way worse in the world of pedal modeling in digital multieffects units.

 

[N. Kentucky Bass]

Sounds like legitimate copy capitalism. Building a business on the backs of people who did the original thinking. No wait -- I wasn't talking about generative AI!

 

[AgingEngineer]

This has led to one thing that endlessly irritates me, though: Euphemisms for brand names in modelers.
My Fender amp has models for "Rumble", "'59 Bassman", and "SWR Redhead".

Line 6 does the same thing, which makes finding a specific model difficult. I've been using the Line 6 Helix for 8 years now, and just last month found out it has a SansAmp model in it. This site is my go-to for decoding Line 6 Helix model names: Line 6 Helix Effect Finder

 

[Wasnex]

Automatic music creation for people that only know that kind of music.

I believe this album was made from samples from old rejects of the record bin. There are some really cool sounding samples, but I think I enjoyed the NPR article on the album more than the album itself:



This one should have been called "Stoned Borg Jive"....IMHO of course.

 

[beans-on-toast]

I'm going to propose a "what if" scenario, but need to explain the backstory first. My amp rig has evolved over the last 30 years and today I'm using the Line 6 Helix for my entire signal chain for live shows, and am very happy with it.

Yet, like any gearhead with GAS, I am always looking at the next cool toy.

A thought struck me: There has been a lot of modeling amps released since the Helix, but it seems like we've hit a plateau. Pretty much any modern modeling amp will give you great tone, so anytime I listen to a new unit, there's no surprises there. So, what's next then?

As a joke --because in my real life I am a systems engineer in the IT / manufacturing space and every single corporation is spending millions scrambling to incorporate some sort of AI to outrun their competitors-- I said to a friend:

"I'm just waiting for someone to introduce an AI modeling amp. You'll train a patch on a specific player's style and tone, then the AI will real-time generate output that sounds like the artist you trained. You could be a complete novice barely pounding out a recognizable riff, and it comes out sounding perfect. On pitch, in time, AI generated riffs and fills in the right places, sounding just like that trained artist no matter how poor the input."

It was funny until it struck me that if anyone thinks anything, that means someone is probably already working on it.

They are doing that with vocals and video now. Sampling the voice and recreating in the style of. New songs by dead artists are appearing.



Compare this to the original version of I don’t Know on Egypt Station (on YT).

 

[ardgedee]

Sounds like legitimate copy capitalism. Building a business on the backs of people who did the original thinking. No wait -- I wasn't talking about generative AI!

What gets protection and what doesn't gets pretty arbitrary. For example: The design of type fonts cannot be copyrighted in the United States. However, the code that programs use to render them can. In different countries in Europe, fonts can be copyrighted (or protected under analogous legal terms) for limited periods of time -- in Germany, after 25 years a font can be freely digitized by anybody and those digitizations can each be copyrighted. In most countries, however, a font's name can be trademarked, which can extend for an indefinite period, so basically even where you're legally permitted to clone Comic Sans, you can't call your copy Comic Sans, it has to be Comical Sans or Cartoon Sans... though not those specifically, they're probably trademarked too.

 

[seamonkey]

This has led to one thing that endlessly irritates me, though: Euphemisms for brand names in modelers.

My Fender amp has models for "Rumble", "'59 Bassman", and "SWR Redhead". Cool so far. And then it has models for "KGB 800", "Monster" and "Rockin' Peg". uh, okay. One's reasonably obvious (although not particularly specific), the others I still haven't figured out.

Unless you're already thoroughly familiar with the original equipment there's no way you'll know what some settings are supposed to be. And it's way way worse in the world of pedal modeling in digital multieffects units.

The current wordplay is to use "Based On" in front of what it's based on.
For example: "Based on the Ibanez TS808"

They may throw in an @R in there in some cases.
"The BASSRIG Super Vintage is, quite simply, a complete vintage bass amp in a pedal. Based on the iconic Ampeg® SVT®..."

Seems to be used everywhere now.

 

[Deleted member 365158]

How about we just stop feeding AI.

 

[c-ba55]

This has already happened. They've done it with singers, so bass would actually be easier. You can sing whatever, and then have it snapped to the midi of what it's supposed to be, with the target signature sound.

 

[logdrum]

As someone who worked on hardware accelerators for AI. (I specifically worked on adders and high bandwidth memory ).. I hate what's it's doing to art. But I'd say that some creative music say the works of avant garde guys like Anthony Braxton , William Parker or Albert Ayler won't be easy. It's not even easy for Berklee or NEC or New Music school grads.

 

[jdjk7]

I'm going to propose a "what if" scenario, but need to explain the backstory first. My amp rig has evolved over the last 30 years and today I'm using the Line 6 Helix for my entire signal chain for live shows, and am very happy with it.

Yet, like any gearhead with GAS, I am always looking at the next cool toy.

A thought struck me: There has been a lot of modeling amps released since the Helix, but it seems like we've hit a plateau. Pretty much any modern modeling amp will give you great tone, so anytime I listen to a new unit, there's no surprises there. So, what's next then?

As a joke --because in my real life I am a systems engineer in the IT / manufacturing space and every single corporation is spending millions scrambling to incorporate some sort of AI to outrun their competitors-- I said to a friend:

"I'm just waiting for someone to introduce an AI modeling amp. You'll train a patch on a specific player's style and tone, then the AI will real-time generate output that sounds like the artist you trained. You could be a complete novice barely pounding out a recognizable riff, and it comes out sounding perfect. On pitch, in time, AI generated riffs and fills in the right places, sounding just like that trained artist no matter how poor the input."

It was funny until it struck me that if anyone thinks anything, that means someone is probably already working on it.

I mean, machine learning is already being used in effects plugins in order to optimize the plugin for a certain sound. But that AI is just adjusting parameters, not creating sound where there wasn't. I don't even really consider that to be a bad use of AI. If music is going to use AI, I'd rather it be as a shortcut to nitty gritty technical tinkering than as a way to subvert creativity.

 

[gtirard]

I mean, machine learning is already being used in effects plugins in order to optimize the plugin for a certain sound. But that AI is just adjusting parameters, not creating sound where there wasn't. I don't even really consider that to be a bad use of AI. If music is going to use AI, I'd rather it be as a shortcut to nitty gritty technical tinkering than as a way to subvert creativity.

Exactly what I think, but we're old chaps, we won't last long. For the new generation, I think music is becoming just entertainment, it can be AI generated, they won't tell the difference. Or maybe there will always be a small market for real art and real music, and a large one for sheer entertainment. Hopefully.

 

[AgingEngineer]

For the new generation, I think music is becoming just entertainment, it can be AI generated, they won't tell the difference.

This is the trend, unfortunately. But then again, that is "unfortunately" from my perspective. I remember when being an audiophile was a thing. When we researched for hours in catalogs and stood in stores comparing the relative merits of the THD of this amplifier and the drivers and solid wood construction of this GIANT TOWER of speakers for our home stereo... just so we could get the best sound when we dropped that needle into the groove. We'd spend weekends upgrading the stereos in our cars, not to make them loud enough to pop a window, but for the most immersive (yes, and loud) quality sound we could get while driving.

For the new generation, music is an internal experience heard through earbuds or played on small portable speakers. They don't think about quality or the ability to hear dynamics. Popular music is mastered to maximize the overall volume because of those listening habits; dynamics went away and instead we have different types of breaks or mix-focus changes in the music so there is a simulation of dynamics.

Multiple studies that quantified thousands of popular songs over the last 50 years show a significant change in how songs and melodies are written. Complex melody morphed into shorter phrases where distinct vocal articulation is valued over melodic content. There is a pronounced shift to shorter hooks with a LOT more repetition.

This is all indicative, and tailored to, of the listening habits of today's listeners. They don't listen to albums, and will punch "Next" on Spotify within seconds if you don't grab them immediately. There's no time for story, just gimme the hook or move on.

But that is a big-market generalization, of course.

If you want to make money, these are the rules to get the streams and be embedded into social media feeds. But, there will always be true music lovers that will march to their own beat. And like history itself, music over time is a living breathing thing. When we hit the plastic wall (which we are pretty much at today) someone, somewhere, will lash out in frustration with a new style, a new sound, and once again the paradigm will shift.

Now, get off my lawn. I gotta get some coffee.

 

[gtirard]

This is the trend, unfortunately. But then again, that is "unfortunately" from my perspective. I remember when being an audiophile was a thing. When we researched for hours in catalogs and stood in stores comparing the relative merits of the THD of this amplifier and the drivers and solid wood construction of this GIANT TOWER of speakers for our home stereo... just so we could get the best sound when we dropped that needle into the groove. We'd spend weekends upgrading the stereos in our cars, not to make them loud enough to pop a window, but for the most immersive (yes, and loud) quality sound we could get while driving.

For the new generation, music is an internal experience heard through earbuds or played on small portable speakers. They don't think about quality or the ability to hear dynamics. Popular music is mastered to maximize the overall volume because of those listening habits; dynamics went away and instead we have different types of breaks or mix-focus changes in the music so there is a simulation of dynamics.

Multiple studies that quantified thousands of popular songs over the last 50 years show a significant change in how songs and melodies are written. Complex melody morphed into shorter phrases where distinct vocal articulation is valued over melodic content. There is a pronounced shift to shorter hooks with a LOT more repetition.

This is all indicative, and tailored to, of the listening habits of today's listeners. They don't listen to albums, and will punch "Next" on Spotify within seconds if you don't grab them immediately. There's no time for story, just gimme the hook or move on.

But that is a big-market generalization, of course.

If you want to make money, these are the rules to get the streams and be embedded into social media feeds. But, there will always be true music lovers that will march to their own beat. And like history itself, music over time is a living breathing thing. When we hit the plastic wall (which we are pretty much at today) someone, somewhere, will lash out in frustration with a new style, a new sound, and once again the paradigm will shift.

Now, get off my lawn. I gotta get some coffee.

Agree on everything, the worst being the kids in the street who "listen" to "music" throught the speakers of their phone. If the new mastering standards is aimed at sounding good on phone speakers, how can we keep any nuances and dynamics?

 

[beans-on-toast]

How about we just stop feeding AI.

It scours the internat public domain. Github is a large public domain software repository. It learns by reading the code. With chat AI apps you can use it as an assistant in developing new code. Start with “I want to write a driver in the style of a github account.” Code will be generated, written in the style of that user account based on what it’s already read. Companies that want their info private are setting up private space versions that nobody outside the company can access. I don’t know how safe this information will be.

PCB design is well suited to AI assistance. Design is done following rules and AI is well suited to that. You provide it with links to data sheets that will be used, specify a board size, etc, and interact with the AI software giving it sub tasks to do.

Note: AI requires a human expert to oversee the design. AI makes mistakes and needs to be told to correct errors. Sometimes you need to iterate to get things right. It then learns and doesn’t make the same mistakes. How well it works depends on the app and the database.

ChatGPT is good and evolving at an alarming rate. They often issue major revision updates. An example of a PCB assistant is flux. All CAD companies are taking this route.

A Better Way to Build PCBs | Flux

 

[Vincent Van Low]

I'm going to propose a "what if" scenario, but need to explain the backstory first. My amp rig has evolved over the last 30 years and today I'm using the Line 6 Helix for my entire signal chain for live shows, and am very happy with it.

Yet, like any gearhead with GAS, I am always looking at the next cool toy.

A thought struck me: There has been a lot of modeling amps released since the Helix, but it seems like we've hit a plateau. Pretty much any modern modeling amp will give you great tone, so anytime I listen to a new unit, there's no surprises there. So, what's next then?

As a joke --because in my real life I am a systems engineer in the IT / manufacturing space and every single corporation is spending millions scrambling to incorporate some sort of AI to outrun their competitors-- I said to a friend:

"I'm just waiting for someone to introduce an AI modeling amp. You'll train a patch on a specific player's style and tone, then the AI will real-time generate output that sounds like the artist you trained. You could be a complete novice barely pounding out a recognizable riff, and it comes out sounding perfect. On pitch, in time, AI generated riffs and fills in the right places, sounding just like that trained artist no matter how poor the input."

It was funny until it struck me that if anyone thinks anything, that means someone is probably already working on it.

just a matter of time....I remember seeing a TV commercial for the World Wide Web in the middle 90's sometime and laughing...who would buy off the computer hahahha and not go to the store hahahahah...

 

[ardgedee]

It scours the internat public domain. Github is a large public domain software repository. It learns by reading the code. With chat AI apps you can use it as an assistant in developing new code. Start with “I want to write a driver in the style of a github account.” Code will be generated, written in the style of that user account based on what it’s already read. Companies that want their info private are setting up private space versions that nobody outside the company can access. I don’t know how safe this information will be.

Very little of public Github is public domain. Public domain means that no ownership is asserted and any use is permissible. Most code on Github is free and open-source code, but with formal licenses that have rules stating what is and is not permissible. Often Apache or MIT, which are liberal in what users are permitted to do with the code but do not allow everything possible, and they still assert clear ownership over the source.

This turns out to be a major problem when using an LLM to generate code: If it's appropriating source code which has been licensed by the Gnu Public License (another of the more popular FOSS licenses, and possibly the most famous, at least within software development), the terms of the GPL encompass all the code that incorporates the GPL-licensed code, meaning that your software is now also GPL-licensed. Which is of no meaningful consequence when you're hacking up a shell script to do a convenient file management task on your personal computer, but it has severe consequences in a corporate environment.

At my last coding job, before ChatGPT existed, we had to have license reviews before every software release, to ensure that we had not incorporated any code that had a GPL or non-free commercial license. Senior developers and lawyers were involved. It was slow, annoying, and had hundreds of millions of dollars at stake (as one of those companies making huge corporate backend software, we were less bothered with the possibility of our code being made public than we were with our clients suddenly being able to use our work without paying us any more). ChatGPT does not cite its sources, so relying on it can be a bad idea with massive consequences.

Companies have been using private code repository services hosted by Github and its competitors for over a decade. It's a common and, as such things go, reasonably successful form of cloud services, and had justified its existence long before ChatGPT came into being. Github, which is owned by Microsoft, which also uses Github as their own private repo, should presumably know better than to allow LLMs to scrape private code, even if Microsoft is perfectly happy to allow LLMs free reign over all the public repos it hosts.

 

[bearhart74]

Would be interesting to have an AI system that could emulate players of all instruments.. Then you could play in a band with Neil Peart on drums, Miles Davis on trumpet, Jimi Hendrix on guitar, Vladimir Horowitz on piano, Sarah Vaughan on vocals and you could be Jamerson on the Bass.