On Machine Learning and A.I. Technology Used for Guitar Modeling

(excerpted from a recent GS post I made...)

The problem with sims to this point is that:

- Syntax - people using generic terms, or misusing terms
- Apples and oranges - comparing different technologies
- Non-linear vs. snapshot comparisons.


Making an IR that sounds identical to a *static* IR is straight forward, but also Occam's Razor invoking. "Why does it sound identical... sometimes?"
Algorithmic approaches feel better, and *seem* realistic but *not identical*. I can adjust my playing to sort of mimic the Real Thing, but "like a real amp" is not the same thing as "just like a real amp". I love Bandmasters because of the variety of timbre you can get from pick attack, dynamics; you can fake that with Certain Emulations, but it's not as subtle and controllable. I have to exaggerate and worse, think about doing it.

That's no fun.

And speaker IRs are only as good as the engineer that made them, what mic and where they put it on what speaker through what preamp at what level. A greenback sounds different cranked; if you play soft it acts completely different, sounds different. Again, there are ways around that, but it's a kludge and it requires conscious effort.

You can make it work "like a real amp", but not *just* like a real amp.

That's not fun.

All of these things I believe can be fixed, but I'm not sure if the programming talent and effort is in the right place, or guided suitably. They're all almost good, suitable for "most applications", but not exactly the same across all applications.

I think the *curation* at companies has been the most important thing until now, the success of some and the downfall of others. ML/a.i. may make it moot, but the "curation" of the training will still be a factor in the end result production-wise. The end result may be fully convincing, but not in the production style desired, which might be a new problem.

/ $.10

What Guitar Should I Buy (redux)

I'm going to try to make an abbreviated version of this very common question. Just about everything new is "basically ok". This has been the case now for a few years.
If you can spend around $250, it's easy. There are 3 "brands" that immediately come to mind, and surprisingly they're not "the expected names".

Monoprice - who is now selling for the same price through Amazon as through their own website. Yes, the HDMI cable company! Their $110 "strat" clone is probably the most cost effective starter guitar there has been. Manufacturing quality is good, potentially a gig-quality guitar.

Harley Benton - a British mail order brand that is available through the Thomann U.S. website. A bit finicky to locate and order, but this line of guitars are well thought out an "curated" clones of popular guitar models. Maybe a bit more expensive than the Monoprice, and a bit longer to have shipped, but with more variety is you choose to study specifically what you want.

EART - Available on Amazon, you can consider these "upscale cost effective".

That's it. You don't have to have an amp with an electric guitar, you can still hear it unplugged. You can also find little rinky-dink $20 amplifiers on Amazon or Ebay if you hunt for them, although alternately you can no doubt find a used practice/starter/beginner amp on Craigslist or at a pawnshop in the $50 range. How's that for brevity?

The Golden Age of Guitar Consumerism

  "... when I was your age, we put chicken wire strung on nails on a rotten 2 x 4 and called it a "guitar", and darn it, we liked it!!!"

 Guitars have never been more affordable.  That's not the same thing as "cheaper" - technically you could buy guitars at one time for $20 from a Sears and Roebuck catalog - but it wasn't very playable.

 

Inside of a $200 Chinese "generic dreadnaught acoustic" 

 There have been different eras in "Guitar Affordability":

 POST LUTE:

 In this era, at the dawn of the creation of "guitar", you only had the choice of getting a handmade guitar.  The relative cost, and quality, can't be quantified, but undoubtably it was more expensive than now.  Strings were made out of animal guts....

 POST INDUSTRIALIZATION:

 Guitars that were manufactured, albeit by hand.  The dawn of companies like Martin, Gibson.  Relative cost depended on quality again, and given that steel strings started around the very early 1900s, people still often substituted "other found wire" because of cost considerations. "Acceptable action" was probably "not acceptable" by today's standards, something to keep in mind when listening to early wire recordings.

 POST JAPAN/KOREA INDUSTRIALIZATION:

  The beginning of machine fabrication on a large scale.  This is where China is now, except they've got modern machines/tools.  Early examples were pretty crude, but playable (and now coveted by masochistically deranged "Cheap Retro Guitar" collectors...)  Later, some of these guitars were pretty good, good enough to make Gibson sue Ibanez because they were making functional Les Paul copies.  They would still go for around $200+, adjusted for inflation many more hundreds.  The cheap Asian guitars could be found for around $100, $50 in a pawn shop - but the reader must keep in mind that's in 70's/80's dollars.  While the Ibanez guitars approached "modern quality and playability", in general these were not guitars one would want to buy today.  

 From the mid-80s onward Japan took over and invented the "modern high quality cheap guitar".  Ibanez, Tokai, Fender Squire (Fuji Gen Gaki) set a new bar.   Korea entered this field with Westone/Electra, Hondo branded guitars, a forgotten transition-period brand that got lost in the coming Ibanez storm. 

 This is when I started playing guitar.  5 years earlier, and my beginner choices would have been dismal.   At the time in the mid 80s, my only choice was made by a company called Hondo, which is effectively what is now known as the Korean Samick brand.  These were not as good as Ibanez, but not as crummy and unplayable as earlier Asian guitars.  Hondo was "the" beginner guitar for a few years until Ibanez steam rolled the field, requiring Fender to step up with the Squier brand.

 POST 80s IBANEZ/SAMICK DOMINATION - THE CHINESE JUGGERNAUT:

 I can remember when China first came on my radar as a guitar builder around .... 2002?  I started to see various corporate-labeled cheap lines with "Made In China" stickers, and they seemed pretty ok.  But they really stepped it up I think around 2010, and as predicted, were going for a transition from OEM to their own branding.  It was obvious they would do that, and while they're not "technically" branding guitars today, it's now a known phenomenon that "China can make good guitars".

POST ALIEXPRESS AMAZON ERA:

 Now.  China has so subtly maneuvered into dominating the guitar market that it's like they've done shades of marketing.  Very quickly from selling clone guitars on Aliexpress - while being OEM for many brands - to "somehow" being used as OEM for alternate brands (Monoprice), and *winning* by making a better product. 
 
 Now there are the Monoprice guitars, EART, Harley Benton, et al - and they're good guitars, and less than what used to be the equivalent entry level guitar.  While being as good as what used to be a "mid level" guitar, or even higher. 

 What you can get on Amazon for $200 from one of these brands now is astounding to me.   Somewhat of what would be considered an "upper tier" guitar in the 80's.  EART and Harley Benton both offer stainless frets, roasted necks and bodies, great hardware/bridges/tuners, good pickups, and they come set up pretty good as well. 

 When Ibanez had to start offering guitars made outside of Japan, that was a watershed moment.  Maybe Samick, who has massive production facilities in Korea, can hang on?  Hard to know during the covid pandemic, with production being dependent on the health of a lot of workers (note China *does not mess around with trying to knock down covid outbreaks...).   Indonesia has tried to enter this battleground, but they've tried to scale against an already optimized market.

NEXT LEVEL - POST PRODUCTION?

 There isn't a next level.  China is at diminishing returns production wise.  But here is "the next level" that nobody has considered:

 1) Guitars don't evaporate.  They go in a corner, in a closet, under a bed, but they don't go away. 

 2) at this point there are probably enough used guitars on the planet that there is probably a positive ratio of them relative to people that even think for a moment they want to play guitar.  Everyone knows more than one person that plays guitar; I'd even suggest "most families" have a guitar at their disposal.

  3) China has relied on manufacturing *growth*.  The situation with guitar is what I'd think is maybe the penultimate last frontier.  They'll have to go after the car market next (which they are... and the U.S. and EU governments will let them crush our own manufacturers with imports...).  But after that, they're at diminishing reeturns GDP.  Solar will still continue to grow, but as efficiencies go up they'll hit a ceiling pretty quickly in 5 years (particularly as demand scales up due to international energy costs/infrastructure problems).   

 The $200 guitar is almost at a point where one can *objectively* say "that's as good as a manufactured guitar can get".   Meanwhile, if you count how many hundreds of millions of guitars have been made to this date, someone really wanting to learn to play guitar can probably ask around and find someone willing to lend them a guitar, or even have one.  A parent today is likely to have one I'd suggest.   

 

Approaching the Uncanny Valley of Guitar Playing

  I've written elsewhere about how a.i./machine learning is going to completely change the musician landscape, as far as production is concerned.

 People do not realize the leap forward we're at right now.  I won't rehash what is available elsewhere, except to say "it's probably not what you imagine it to be".   What is happening with Pytorch/Magenta/Deep Mind etc.  is as big as the Internet was, and will change us and our lives as much as it has. 

 I recently tried the Magenta project's plugin.  The sax version is amazing; it does what I promised would one day happen with ML/a.i. software, in that it doesn't just make the input come out having the timbral sound of "a saxophone", but also the inflections.

  To the reader: if you think what I'm talking about is akin to playing a synthesizer with a saxophone patch, you're wrong.  It is a completely different thing.  The program is doing things the developers literally don't totally understand; it is working on a pure dataset level.  It is NOT a DSP based wave shaping technology.

 It's a bit tricky to handle.  I have to imagine saxophone playing to match the dynamics, attack and vibrato - but when you get it right it's uncanny.  I wish the violin version worked as well, it's very creatively liberating: as far as I'm concerned, I can add a "tenor sax" part to a recording.

  I presume there will be a guitar equivalent soon.  You'll be able to whistle and have it come out with an inflected guitar sound.  The question is in the quality of the model training (that governs the output); as I predicted, soon it will be possible to make a training model create an output that "corrects" a player's dynamics and inflection to sound equivalent to "Stevie Ray Vaughn", "Brian May", etc...

 Unlike the saxophone, though, the variety of guitar sounds I think will make it impossible to have it be flexible enough to cover "all" styles.  And using it will pigeonhole your choices into a certain way of playing, just as the "saxophone" plugin does.  You can't "play" the saxophone plugin like guitar, and have a good result.

 So in the future - people will specialize in how they operate with their ML sounds - after a period of people getting confused/impressed by the realization of the technology actually working.  A period equivalent to "keyboard popped octave bass sample basslines", and then a maturity.

 One downside is that it will making mocking up a cliche clone of a Known Famous Song very easy, and many will do it and garner kudos for it.  The confusing effect of this is going to be a big negative.

 Another potential negative is - a company will jump on this to have it in a guitar amp.  I've been saying this for awhile: a beginner amp with this technology can have a couple of presets that will not only yield an output that sounds just like the original recording of a Famous Player, but correct dynamics and probably pitch as well.   Harmony will be a problem I think for a few years, but "lead guitar playing" is about to undergo a disaster in that people will think even less of the skillset required to ACTUALLY DO A GUITAR SOLO.   

 A renaissance mentality will maybe become a trendy thing, hopefully.