The No.1 Website for Pro Audio
MPE, machine learning and increased processing
Old 29th May 2019
  #1
Gear Maniac
 
Muied Lumens's Avatar
MPE, machine learning and increased processing

Good times are ahead. Or bad, considering the possibilities that probably will dumb down the music creation process as seen from today's perspective. Personally, I'm excited about the future, and what seems to be brewing in technology in general. In fact, I believe we haven't even started yet!

I remember Brian Eno once musing on a synth with only two buttons: one which you press when it does something you like, and one when you don't. Whether this exact tech will actually materialise is irrelevant, but I believe he had a point. The simplicity of the interface will appeal to many people, without sacrifising the flexibility of something that becomes unique to only you.

Machine learning, or AI, or whatever you want to call it, will have the ability to learn from your behaviour and get at least a basic understanding of what your preferences are, so in effect what we could get is something along the lines of an advanced arranger keyboard as your jammimg buddy. (I know, I'm whincing as I write it...)

What you could do is concentrate on the overall feel of the music, becoming more of a conductor than a composer, but with at least some control over everything from the timbre to the tempo. So what you could do (at least to save your ego) is learn how to play propely, using an MPE controller with highly expressive synthesis techniques, in a steady stream of musical output that hardly anyone will hear in the noise of the future internet. Well, that last bit is already here, at least for me.

This is not me trying to predict the future - it's already here. Think Big Data and gadgets such as Alexa. At least in a small way. Take the MicroFreak for example; it has a randomisation feature in its sequencer. Maybe it is a sign of things to come, or just another fad, but think of it this way - if it eerily came to know your preferences as you used it more, would the novelty wear off? It would develop along with you, as your tastes change and your abilities improve.

The tech is here, at least in broad strokes, but people might reject it. I think I would, if what my synth is doing is based on what 1 million other people prefer. I hope that doesn't happen.
Old 29th May 2019
  #2
Lives for gear
 

Yeah I've probably got various things to say about this but lacking time right now.

So for now I'll just say that the visual applications of machine learning etc seem to be where the bulk of the action is right now, and it is intereting to try to extrapoloate stuff they are doing in those realms with what could be applied to sound and music one day. I'm thinking primarily of short animated experiments I keep seeing on twitter etc where a painting or photo of someone can be combined with a video of someone else, and the face from the painting or photo is transplanted and comes alive with the movements from the video face.
Old 29th May 2019
  #3
Lives for gear
 

Plastic music for plastic people- it's been here for a while.
Old 29th May 2019
  #4
Lives for gear
 

Quote:
Originally Posted by Treebase DMX View Post
Plastic music for plastic people- it's been here for a while.
Much as some people will have cause to find some mainstream stuff too safe, cheesy or similar, and much as the same tired stories and sentiments may be overrepresented in film, music, tv etc, we remain a long way away from the 'machine created prole food' that Orwell spoke of.

Yes, from certain valid but bleak angles we can have valid conversations about all sorts of plastic, soul-less, automated pap. But its never the whole story, and humans do retain an interesting sense of what is authentic. People dont mind the blend of human and machine as long as something human remains, you can have authenticity where it matters no matter how robotic the tools.

I'd rather think of a lot of the AI and machine learning stuff in the sense of it just being more sophisticated tools. Tools that do have the potential for unintended consequences (and certain unloved but intended ones), but away from the headline grabbing worst-case dystopian nightmare scenarios, plenty of other stuff can and will happen.

For a start, some of this stuff has the potential to free us from some of the existing technology which is crude and confining. The fixed grid has many uses and changed the game in a number of ways, but better tools that let the grid bend could be something that increasingly happens and enables future technology to eliminate one of the downsides that old-current technology brought to the world of music.
Old 29th May 2019
  #5
Lives for gear
 

If I train a model to forecast umbrella sales, but don't feed it weather data, it's going to be a pretty crappy at its job.

Same is true feeding an algorithm only point data regarding playing style, lacking context that makes music magic.

We're pretty far away from the singularity. Unless they won and this is a simulation because batteries?
Old 29th May 2019
  #6
Gear Maniac
 
Muied Lumens's Avatar
Quote:
Originally Posted by SteveElbows View Post
For a start, some of this stuff has the potential to free us from some of the existing technology which is crude and confining. The fixed grid has many uses and changed the game in a number of ways, but better tools that let the grid bend could be something that increasingly happens and enables future technology to eliminate one of the downsides that old-current technology brought to the world of music.
Exactly. I also think there might be some potential in synthesis, where not just a huge amount of parameters can be adjusted by training methods, but new physical models can be created based on the sounds you prefer. I know, I'm dreaming a bit now...

Or FM patches which are created based on input that you feed it, like resynthesis but with phase modulated oscillators, which you can then go in and mess about with... OK, I'll stop, for now.
Old 29th May 2019
  #7
Gear Maniac
 
Muied Lumens's Avatar
Quote:
Originally Posted by lost_the_peace View Post
If I train a model to forecast umbrella sales, but don't feed it weather data, it's going to be a pretty crappy at its job.

Same is true feeding an algorithm only point data regarding playing style, lacking context that makes music magic.

We're pretty far away from the singularity. Unless they won and this is a simulation because batteries?
I'm not talking about self aware groove boxes here, just things that exist now, which might - or might not - trickle down into the world of music technology eventually. I might be decades, even.

But otherwise you might have a point.
Old 29th May 2019
  #8
Lives for gear
 

Quote:
Originally Posted by Muied Lumens View Post
I'm not talking about self aware groove boxes here, just things that exist now, which might - or might not - trickle down into the world of music technology eventually. I might be decades, even.

But otherwise you might have a point.
Fair cop!

I think it would be cool to have an AI buddy vst that can paraphrase what I'm playing in the style of Bach so we can jam along. So long as the vst is called "Bach to the future".
Old 29th May 2019
  #9
Lives for gear
 

Quote:
Originally Posted by Muied Lumens View Post
Exactly. I also think there might be some potential in synthesis, where not just a huge amount of parameters can be adjusted by training methods, but new physical models can be created based on the sounds you prefer. I know, I'm dreaming a bit now...

Or FM patches which are created based on input that you feed it, like resynthesis but with phase modulated oscillators, which you can then go in and mess about with... OK, I'll stop, for now.
Yeah. I need to get much further along in my general synthesis journey before I can properly contemplate all the things on that front.

I do have a long term MPE obsession and am currently experimenting with using x,y,z positional data from VR hand controllers to control synth parameters in a monophonic and duophonic expressive manner, as opposed to the polyphony of MPE. I've only just started this, having previously tried and quickly given up on something similar with the Kinect many years ago.

One reason I focus on this stuff is that development and gluing and mutating other peoples software/hardware is sort of my thing. I'm usually not clever enough to do that hardcore programming at the heart of fancy systems, but I am good at standing on the shoulders of giants and welding other peoples stuff together in different ways. And I know my way around Unity game engine, so I've used that to transmit OSC messages from an Oculus Quest to a computer running VCV rack with a OSC->CV module loaded. And I have a DC-coupled audio interface so I've been able to send that cv to eurorack and Moog Grandmother so far. This is about my level, slightly beyond just using eurorack modules, but some way short of being enough of a developer to actually make any modules of my very own.

So yeah no surprise that my brain has usually gone foggy on me when I've actually attempted to understand machine learning and AI stuff from a programming and development perspective in the past, but I also lacked incentive and a particular application for it at the time. Since Unity has some AI/machine learning stuff in it these days, I might try to learn a bit of that to see if the concepts sink in this time, and then turn an understanding of this towards musical realms. I'm not good with chords for a start, and I dont suffer from many stigmas about 'cheating' on this front since I'm not a proper player of any instrument either, or any lofty aspirations. I want to put most of my attention into the expressiveness of my playing rather than some of the other hard work that normally goes into things, often from an age far far younger than I found myself starting from.
Old 29th May 2019
  #10
Lives for gear
 

On that last point, for several decades now there has been some evidence that I have an especially awful memory of the sort that is most used to store the entire sequence of notes, chords, lyrics and all the things you need to know to play them. I dont know how to describe this properly, but it is one of my impediments to learning an instrument, even when I take the time to get physically comfortable with the act of playing it, my mind goes very blank when it comes to what notes I am playing. There may yet be techniques out there that help me get past or round this in some key way, but in the absence of discovering a solution to this issue of mine, I would like tools that can assist me in ways that go beyond the rather manual and simplistic helpers of today. Dont get me wrong, even some simple midi fx can do a lot if you take the time to learnt he theory of what you are trying to do musically, there are already many performance crutches available or ways to avoid that sort of performance altogether. But I'm certainly interested in greather and more sophisticated blends of live performer and AI tools/helpers in future.
Old 30th May 2019
  #11
Gear Maniac
 
Muied Lumens's Avatar
Quote:
Originally Posted by SteveElbows View Post
...But I'm certainly interested in greather and more sophisticated blends of live performer and AI tools/helpers in future.
Me too. Predictive texting gone jazz.
Old 30th May 2019
  #12
Quote:
Originally Posted by Muied Lumens View Post
....

Machine learning, or AI, or whatever you want to call it, will have the ability to learn from your behaviour and get at least a basic understanding of what your preferences are, so in effect what we could get is something along the lines of an advanced arranger keyboard as your jammimg buddy. (I know, I'm whincing as I write it...)
Why would anyone have someone else make music for them?
The enjoyment and skill is in creating, playing it. Especially playing together with other musicians.
Quote:
Originally Posted by Muied Lumens View Post
The tech is here, at least in broad strokes, but people might reject it. I think I would, if what my synth is doing is based on what 1 million other people prefer. I hope that doesn't happen.
Without trying to steer this discussion into political water, music creation is not the priority function that AI are now being developed for. As far as I know we don't have "true AI" right now. There are rational reasons to reject the technology completely.
Old 30th May 2019
  #13
Lives for gear
 

Quote:
Originally Posted by SteveElbows View Post
On that last point, for several decades now there has been some evidence that I have an especially awful memory of the sort that is most used to store the entire sequence of notes, chords, lyrics and all the things you need to know to play them..
Did you get much music education as a child? I think adult learners tend to be at a serious disadvantage when it comes to this.
Old 30th May 2019
  #14
Gear Maniac
 
Muied Lumens's Avatar
Quote:
Originally Posted by Reptil View Post
Why would anyone have someone else make music for them?
The enjoyment and skill is in creating, playing it. Especially playing together with other musicians.
Yeah, good point. And playing with others will never go away. If it ever happens, this is just another tool for musicians.

You are speaking from the point of view of a musician, though. Perhaps the lines between creators and consumers will blur even more in the future than they do now. 'Enjoyment' is not easy to pin down, but if it is easily won, it could also be easily lost. That could mean a further depreciation of any form of creation, not just music, which could be problematic in many ways.

There are two things I believe will always stay constant though, even though they contradict each other:

1. People with talent will always shine when they work hard and know how to make use of their skills.

2. It's not always the best person who gets the gig.

Any sort of new technology won't change that, at least in the foreseeable future.

Quote:
Without trying to steer this discussion into political water, music creation is not the priority function that AI are now being developed for. As far as I know we don't have "true AI" right now. There are rational reasons to reject the technology completely.
Again, I'm not going that far in to Terminator territory, I am talking about what seems to be just out of reach currently. If you want, you can try practical machine learning yourself today, by downloading Wekinator: www.wekinator.org

There are also a few options if you are a programmer, like Unity mentioned above, and other tools for many platforms.

Sure, it's early days still and I may not have a clue what I'm talking about too, but that's partly why I posted this thread.

If it doesn't come across clearly, let me assure you that I am definitely in two minds about all this. Data collection by companies is a hot issue right now, as Big Data analytics is starting to take hold. The connection to music technology is not straight forward, but one thing is for sure: we live in interesting times.
Old 30th May 2019
  #15
Gear Maniac
 
Muied Lumens's Avatar
Thanks for moving the thread, btw. It's probably better off here.

Some times I forget that there is more to this place than the Electronic music forum.
Old 30th May 2019
  #16
Gear Addict
 

There will definitely be some better powerful tools in the future that can help musicians work better and can generate some of the various elements that go into music. But I think humans will still need to sort through the results, curate, edit, and ultimately put everything together. If the end result leads to better music, it will be a net positive for everyone.

But fully self-contained music generating AI that can make great music in a wide variety of styles that is indistinguishable from great human-made music is a long way off I think. And where is the creativity? You can use machine learning to train a system on a particular set of source material but it won't have any way of creating something truly new and different beyond that source material. Unless you just add some element of randomness, but that's not really creativity.

I also think that it's very likely that some existing technologies will continue to develop but will hit a wall at some point. If you look those AI generated images that are popular online, they have a recognizable look about them. I wouldn't be surprised if AI synthesis would be similar. It might create some interesting sounds but they might have a certain similar quality about them, or particular artifacts that would be recognizable as coming from the AI. Kind of like how when synthesizers first appeared people were convinced that they could mimic any instrument, but while we got close, it didn't really turn out that way.

Somebody posted this AI generated death metal stream in a recent thread https://www.youtube.com/watch?v=CNNmBtNcccE
Initially it's pretty impressive. But then I found a similar video they did, training the AI on the Beatles catalog. https://www.youtube.com/watch?v=exXUPJRZKZ0
You can immediately see why they chose to do the death metal version. This video shows the progression of the system learning over multiple generations and it gets pretty interesting toward the end. But this is still ultimately just samples of someone else's music and there's no real creativity happening.
Old 30th May 2019
  #17
Gear Maniac
 
Muied Lumens's Avatar
The "AI" music creation is fine, but what if you could do the same with synthesis patches? Could be some potential there, especially when it comes to complex synthesis engines like physical modelling or 64-op FM synths, for example... I mean, presets all sound the same until two different people play them.
Old 2nd June 2019
  #18
Gear Guru
 

Quote:
Originally Posted by lost_the_peace View Post
If I train a model to forecast umbrella sales, but don't feed it weather data, it's going to be a pretty crappy at its job.
Not if the goal is to sell umbrellas rather than forecast the weather.

It might seem like a silly 'retort', but I really think people miss out on the importance of that distinction.

In our 'system' we often create things for the purpose of a particular outcome, in the above example making more money. We don't give a crap about the actual weather and whether or not people get wet, we just care about making money. So if the model is trained according to previous sales and can accurately predict future sales, then that's all we care about and our job is done.

Quote:
Originally Posted by lost_the_peace View Post
Same is true feeding an algorithm only point data regarding playing style, lacking context that makes music magic.
I think the larger issue is just what "magic" is. I would argue that there's a lot of music out there already that lacks "magic", yet people consume it. So feeding current data might indeed produce more magic-less music, but just like above; so what if it's consumed?

"So what" is of course from the standpoint of the makers/owners of the AI, not from my standpoint. From my standpoint we'd be dumbing down society and reducing music as an artform.

Quote:
Originally Posted by lost_the_peace View Post
We're pretty far away from the singularity.
Well, we don't know that. I think the field of AI very well could develop exponentially rather than linearly, and when that starts to happen we're woefully unequipped to respond...
Old 2nd June 2019
  #19
Gear Guru
 

Quote:
Originally Posted by Muied Lumens View Post
Yeah, good point. And playing with others will never go away. If it ever happens, this is just another tool for musicians.

You are speaking from the point of view of a musician, though. Perhaps the lines between creators and consumers will blur even more in the future than they do now. 'Enjoyment' is not easy to pin down, but if it is easily won, it could also be easily lost. That could mean a further depreciation of any form of creation, not just music, which could be problematic in many ways.
I was going to make the very last point.

I think the fundamental question we should ask ourselves is what music is, what art is, and what creativity is. Am I a great drummer if I place notes on a grid and have the computer play that back for me? Of course not. So what am I? A composer? Possibly. But what if the computer keeps giving me suggestions based on AI that I accept? Am I still a composer? 100% me or divided between me and the AI?

The more of the work our tools do the less of that work we do. By definition we are being less creative in any one given task the more someone or something else is doing that task.

Quote:
Originally Posted by Muied Lumens View Post
There are two things I believe will always stay constant though, even though they contradict each other:

1. People with talent will always shine when they work hard and know how to make use of their skills.

2. It's not always the best person who gets the gig.

Any sort of new technology won't change that, at least in the foreseeable future.
Well, we can quibble over "foreseeable", but one consequence of all of this I think will be AI generated and protected works. That is, if someone creates an AI that spits out composition after composition, and all of that work receives IP rights, where does that leave composers in the capitalist market? How will composers navigate making money off of their work when "all has been done" already? AI will only speed this up.

I do agree however that live music in a sense will be "protected" as an art form for a long time since it will take quite some time before actual robots will look human enough to fool us. So a local bar with truly live music performed by human beings will likely be a 'thing' for a long time. But we're in for some big and mostly terrible change in my opinion.
Old 4th June 2019
  #20
Gear Maniac
 
Muied Lumens's Avatar
OK, so composition is one thing, but what about highly expressive presets? Is that too contrived? Perhaps I need to post this in a developer's forum instead...
Old 12th September 2019
  #21
Lives for gear
 
monomer's Avatar
 

In another (unrelated) thread @ Treebase DMX said that machine learning is not learning at all because it is done on a computer by using instructions.
@ lost_the_peace then said that with machine learning a computer can be used 'to learn' but that that is not 'learning'.

I wholeheartedly disagree with this. Learning is learning.

Discuss!
Old 12th September 2019
  #22
Gear Guru
 

No.
Old 12th September 2019
  #23
Lives for gear
 
monomer's Avatar
 

Quote:
Originally Posted by mattiasnyc View Post
No.
Old 12th September 2019
  #24
Lives for gear
 
monomer's Avatar
 

Quote:
Originally Posted by Treebase DMX View Post
Firstly, not everything that is real can be "proven" .
If its real at least you should be able to show its effect on something in this universe, right?

Quote:
I'm not saying AI tech doesn't work.
You literally stated that computers can't really learn because they use instructions.

Quote:
I'm reacting to the language being used and to what it will undoubtedly be used for. Most people who are not tech savvy are already massively victimised and exploited by technocracy and this will escalate beyond belief when all that AI stuff takes off. If you think it's going to make the world a better place, you may be in for some rude awakenings.
Not sure what you mean by language in this context, but i also don't like the sort of power that AI will give certain people and organizations. But that has nothing to do with learning vs 'learning'.

Quote:
Yes, you can make a machine that can learn to do basic tasks but it is really a simulation, a mechanism as you say.
Why do you think that humans don't work based on mechanisms and why is that more 'learning' than what computers do?
Quote:
Living beings are not mechanisms ...
Yes, yes, we are. We are highly advanced molecular mechanisms running on entropy.

Quote:
... and the human mind is not analogous to a computer processing "data"....
In some ways it is, in other ways it isn't.
All analogies break down at some point.
But they certainly are information processing mechanisms. Extremely so even. The brain is all about information.
And if we reproduce (simulate) those mechanisms it turns out the simulations start to show behavior that we can also find in nature.
Of course, nature has had millions of years to fine tune all kinds of information processing mechanisms, starting out with DNA's influence on the environment, but that doesn't mean it's not essentially the same stuff, the same sort of fabric of information processing.

Quote:
True learning is not memorising or finding what works via trial and error...
Aah, here we go. FYI, you've changed subject now. :D It's not about 'learning', it's about 'True learning', something that apparently only humans can do.
So all animals that learn by trial and error or memorization are not capable of learning? Do you realize that trial and error and memorization is done by all known lifeforms as a process to gain information about their environment?

Quote:
Yeah I understand about emergent behaviour and that with enough complexity you could see something that appears to be spontaneously doing something on its own.. but that's not how conscious life works
Who's talking about consciousness? Why is it required to 'learn'? Is a bacteria conscious because it can learn?
Quote:
and the idea that science has figured it all out is just more hubris and vanity
I don't think any scientist is claiming their models are conscious.
Topic:
Post Reply

Welcome to the Gearslutz Pro Audio Community!

Registration benefits include:
  • The ability to reply to and create new discussions
  • Access to members-only giveaways & competitions
  • Interact with VIP industry experts in our guest Q&As
  • Access to members-only sub forum discussions
  • Access to members-only Chat Room
  • Get INSTANT ACCESS to the world's best private pro audio Classifieds for only USD $20/year
  • Promote your eBay auctions and Reverb.com listings for free
  • Remove this message!
You need an account to post a reply. Create a username and password below and an account will be created and your post entered.


 
 
Slide to join now Processing…
Thread Tools
Search this Thread
Search this Thread:

Advanced Search
Forum Jump
Forum Jump