Soundpaint Ramble 002 AI Video Response
Вставка
- Опубліковано 4 кві 2024
- A Soundp(AI)nt Ramble. Let's hear from YOU in A.I. music-making!
- - - - - -
Join us on Discord / discord
Soundpaint™ is a new analog-style sampling technology.
Soundpaint renders all instruments in real-time in 127 dynamic velocity layers. It shortens the distance from idea to realization. Instantly morph a waterdrop with a violin. Play the most iconic instruments of all time in exquisite realism and fidelity. True analog-modeled effects. Advanced and organic arpeggiator. A deep matrix for those who desire.
Welcome to the dawn of Real-Time Sampling™
Soundpaint is available at www.soundpaint.com - Фільми й анімація
Imperfection! Around 1987 I did an arrangement of Here’s That Rainy Day on a music program on the Amiga computer. I demoed the program and played the song for the Sacramento Amiga Users Group one evening at our monthly meeting.
It was well received. But, one member came up to me and said “It was too perfect.”
That comment has never left this almost 80-year old lifelong musician’s mind.
BTW… I think I am going to have to buy the fire horns. So much fun!
The fact that you offer us the same tools you work with is by far one of the best things about you and your team. It doesn’t mean we can automatically become great composers but it is an important way to make apprentices of us all.
I believe AI to be a tool that can help us create faster but it will never supplant human experience.
I reflect back to the prominence of drum machines and the alarm at that time from drummers. The actual result was a new type of music and some hybrid types. Think how Phil Collins - a noted drummer, used a drum machine instead of rejecting it on "In The Air Tonight" or Prince used a Linn to shape his sound and (from what I've heard) drive his drummer to become more technically skilled. These artists and many more used the new tool to create new things never heard before, but still very human and art-based in feel. There is a reason why "humanize" exists in DAW's and drum machines.
Great Troels ! You have summarized exactly what I've been saying for days. Excellent analysis as always.
When I hear the argument that AI can never make music like humans, because it can't comprehend the lived experience of being a human-to know the pain and joy and the everything of being alive-which influences our music, my heart says, “Yes, BUT...” When I look at much of the music on the pop charts, it honestly feels more robotic than human. Trite lyrics, cliched ideas. Sometimes it feels like we're doing the work of the AI for it. So when we approach our music, it's a good reminder to strive to be genuine, open and honest, and to get to the heart of what we're singing about and what we're creating.
Thank you for this video, watching on a Saturday morning with a coffee. To me, a great way to start the day.
Yeah. There are certainly styles of pop music which tends to be more "cardboard" than others. Though I often find myself captured by the production techniques applied. Like high-grade cardboard. But perhaps more importantly is the creation process itself. Clicking a bottom may create something, but the 4 years of grueling self-torture that Beethoven had to endure writing his 5th won't happen. How will an AI truly understand the art of silence. Why the conductor may choose to let a note linger longer. I often think of what made the great masters so prolific. And I think a part of the answer was restriction. All subjects I suspect are of interest to you, hence your name. Cheers, T (about to get my morning cappuccino - and just like Beethoven I do count my coffee beans too).
Thank you for engaging on this, @soundpaintmusic. Your "bottom-up" approach will be key! We human creators will need to "get our hands on" the basic elements in order to produce works with the cohesion (your excellent word) that we yearn for. Then, AI can become our assistant or creative partner. Top-down training doesn't yield on-ramps for people to dance with the AI toolsets. There are too few handles.
Well said Dave. I would almost state it’s musically arrogant … to think an AI will just magically understand what makes Yo-Yo Ma such a great performer. What makes his rendition of Bach so gorgeously unique? Is it him? His background and decades of sharpening his skills? Is it his 1703 Stradivari? How come every single bow stroke is uniquely sounding?
It may seem like an AI should be able to capture a wood box with 4 strings with ease, but … in that “seem” lies to ignorance. Not fully being able to digest just how complicated it really is … Never mind Bachs writing too for that matter. ❤️
@@Soundpaintmusic The issue is: the AI isn’t building the instrument; it’s just building the waveform (at this point). It’s the ultimate subtractive synthesis; carve the sound out of everything that is not the sound. It neither knows nor cares what a cello is. Yo-Yo, on the other hand, knows how the tree grew, how it was sculpted into an instrument, how the string tension torques the neck… And he knows how he grew his body and mind to fit it.
I think that @@davidskybrody has a good point to A.I. doesn’t have the language to create the subtle sound differences with certain musical instruments. I would however say that Yo-Yo wouldn’t have the slightest connection to the tree that provided the wood for his cello. What he would understand in the connection with how the subtle differences of pressure, speed and angle of his bow across the strings will resonate with this particular cello to produce the sounds that invoke the feelings they do. It’s more of a connection and understanding of emotions and having the presence to be inside and one with the music so that you are part of the sound.
A.I. as it currently stands doesn’t understand emotion so it cannot paint those feelings into what it produces. It doesn’t understand the connections between emotion like how you can cry from being so happy and how it is different from crying when you are sad.
There is an energy that we feel at a level that we may not be conscious of. THAT energy is what we feel in music, art, etc. created by a human being. We feel that at many levels but I do not feel that in anything created by AI. I SEE amazing AI illustrations....I never feel them in a way I feel when created by humans.
Agreed. I often think of movies - when everything is just working. Nothing seemingly standing out - it’s just perfection across all the disciplines - and when things stand out they do it with such nerve and intention - that it requires humans in every chain of the creation.
AI will become a tool in that creation process, but not a replacement. We, humans, are far too (im)perfect to be deducted to data driven emulations.
In its ideal form AI will amplify our expression and facilitate more original ways of expressing ourselves.
But if we can all just click a button - we will all be saying the same. ❤️
Aren't we, as humans, sampling as well? We can be taught chords, scales, keys, melodies, etc., but the first songs we learn to play we're created by others. Once we master an instrument we can then begin to create. AI is very very young and it's early on, really. As we listen to music, we find specific genres that we like and then we start to create something in that genre. Just thinking... Always a fun thought process
Good arguments! Allow to 🏓 back.
Let’s agree that AI and humans aren’t the same. We have always for humans and human behavior. We do not have laws for AI and AI behavior. The fact we keep comparing the two only speaks to the need for lawmaking within AI - as it does have many human traits and obviously ability to augment humans.
But as of right now - companies are freely using AIs and training/sampling them on other people’s music. Claiming they don’t have to disclose what music they trained on and with no liability for the training.
Yet they can make commercial services out of it and essentially modify other artists works and create new derivatives in it.
To me that’s the same as sampling music. The simple equation being … An AI cannot generate music without listening to music. It is sampling the data - and that should fall under the law just like sampling someone else’s music does.
No music to sample from. No music coming out of an AI. Same thing. Different packaging.
@@Soundpaintmusic Completely agree - that is where we are going wrong - monetizing a child. As humans, I think it just makes sense to train AI on human things - however, monetizing on that, such as in the creation of music is just wrong.
Agree with that. I consider myself trained on music… And when I compose, I’m not sampling, I’m just using my own personal dataset (call it culture?) which is obviously inspired by other musicians’ creations.
Music is a language, and as such it is a requirement to learn the language if we want to express ourselves. The same language can be poetic, mundane, beautiful, dissonant and often the take on which is what is personal. The truth is no matter how creative we feel we are, everyone quotes their sources, even if subconsciously.
EXCELLENT discussion! One note: At the end you first agreed that, from a music consumption perspective, the way music is produced is not of much concern, but then later conflated that with your own emotions regarding the significance of the production part to you. ...
While so very important, this was not the point of the commenter. The point is is that no matter how rubbish AI art or AI music is, the average consumer will gobble it up for many reasons, primarily... AI allows the masses to create... period... in areas where they may have felt no skill or opportunity to create before. ...
Unfortunately, personalized AI-generated art of all forms will dominate for that very reason - personalized creation - even if deep human emotion and appreciation of music production is excluded from the equation. ...
I feel we all need to begin pushing now for a trifurcation in labeling and perhaps platforming: 100% human-generated, human-AI co-produced, 100% AI-generated. Otherwise, the combined pool will become overwhelmingly saturated with 100% AI-generated works, with very little s[ace, eye or ear-time for human-generated works. Even then, people will be more inclined to produce their own personalized works and share within their circles rather than seeking music outside of their circles. They just won't have the available time nor attention span to do so. ...
With that realization, and all of that said, I still just purchased EIGHT SoundPaint libraries over the last four days. 😂I can't help but be compelled to create my own music, and these libraries compel me to even higher heights! Kudos to the entire SoundPaint team!
Thank you for your observations. I don't think the receiver necessarily cares about how things were made. Music consumption today is much less focused then it used to be. However I do think the creation process matters, since its where we make the h(ear)t speak and get to form our independent expressions. It may not be that the end-listener can discern this consciously, but I do think its picked up subconsciously. The same with production quality in music. People may not be aware of it, but it still matters. Something like that! : ...)
Another thing we should perhaps discuss is why we make music. I personally only make music for- and to myself at this point. If others enjoy it great, but its the exploration process of diving in the soundwaves that makes me happy.
Also thanks for supporting us! Hope the paint and colors inspires your creation process too. Cheers, T!
♥
I used chatGPT to assist me with writing a few poems for my loved ones with excellent results. But it was collaborative. I set the premise and then manipulated the output, changing words and rhythms until I was happy. Yes, the initial output was cheesy and plasticky but after changing and swapping words, I was really pleased with what was produced. I added the ‘soul’. If I use AI with my music, I will do so collaboratively.
However, I think people will be lazy and just ‘push a button’, expecting the output to be just a good as the greats.
Exactly and a good example of collaborative creativity. How neat will it be once we get AI based instruments that will work in similar ways - and ideally not plasticky out of the 📦
Hearing you talk about vocaloid and synthV in this channel was something I never even dreamed of expecting HAHA
I'm a regular user of Soundpaint AND Vocal Synth user, I've been using Vocaloid for a decade, and SynthV a few years ago (SOLARIA REPRESENT)
I think about humanity within artifice alot, this is a theme that I tend to return to in my work.
Basically, i like to implement human imperfection to entities built for perfection.
You're totally right about SynthV getting us 80% there, but thankfully SynthV can be as nuanced as we like it to be. I've done vocal singing studies to understand the microtonal pitch line, timbre change and inflections of human singing and implement my studies into the parameters of SynthV.
The main cool thing I wanna state about Vocal synths is that like real singers, listeners would eventually "perceive" a tuning style affiliated to producers who extensively tune their vocal synths.
Cause there's always a near infinite way to sing, same as there's a near infinite way to tune.
Thank you for not dismissing vocal synths, people in the sphere is so used to their dismissal it's a surprise seeing someone in the industry who can understand value in vocal synths, the whole video in itself was a really fascinating watch
Great points! I think Synth V actually demonstrate how to do AI responsibly! It’s not gonna replace singers, but the whole “Vocaloid” is a new aesthetic of its own and has its whole own sub genre of Hatsune Miku style music.
It gave birth to something new - rather than to plagiarize what’s already there. 🫶🏻🫶🏻
AI isn't going away. It's the future like it or not. I love experimenting with AI tools and using AI to write songs as a musician (hobbyist). It's fun to hear what it comes up with and you can always add your own playing to AI songs which is also fun. Or, take an AI song and remix for humans. All sorts of ways to create new music.
Troels, you have been involved with sampling in legitimate ways and shared your efforts for years. I hope that you could try to share your views on AI as an expert with the lawmakers who are curretnly in the process for creating the rules for it' use.
Thanks Wilhelm. I am involved in a couple of initiates. But would wholeheartedly go to lawmakers and break it down. The tech is here to stay. But it’s like early days of Napster. We know people want it. But let’s do it the right way that doesn’t erode what’s left of the industry. ❤️
AI is clever but human made can also be clever and has inspiration. Inspiration and Imagination. Something AI can't reach.
It’s gonna be an amending creativity tool, especially when it’s creatives (and not just tech people) building- and training them.
Hey Soundpaint Team, it would be awesome if you guys also start a Music Theory series. Your demos are always amazing. And also with the likes of Troels, Natalie and Chimy it would be awesome to see them offer their tips and tricks for composing. Maybe even some Track from Scratch videos with Soundpaint and 8Dio libraries. Let me know what you guys think?
Love it and something we are thinking about. In terms of music theory I am the worst at it (Troels), since I never had a music teacher and just learned by h(ear)t. But I can certainly offer some sample theory or mockup theory lessons. But also have others cover the theoretical stuff. Great suggestion. 🎹🎛
@@Soundpaintmusic amazing. And I think that's the beauty of it Troels. I'm sure our community has people who also have learnt music all by themselves, including me. I always find your chord progressions and musicality really unique and fresh so it'll be really amazing to hear from you too about how you go about it 😄 can't wait for those videos ❤️
Noted!
Your points are well taken. Why would we dis the muse and the gift of creating by abdicating our responsibilities? The legality of sampling is still a gray area, even after all this time, and the data sets should be made public. The other issue is the technology isn't intelligent, it is merely assembling by instruction from said datasets. To me, physical modeling is more AI than AI, but neither are intelligent, they are highly specialized computing. A key is pressed, the computer considers all other keys pressed and their velocity, open strings, the container resonance...And math creates an instrument. Sampling a singer or a style and using it to train a silicon clone adds nothing to the musical lexicon. I can see all of that as useful for a songwriter who needs a scratch vocal for a demo, but it won't replace a singer or bring unexpected nuance to a performance.
However, and I hope this doesn't make me devil's advocate, but a good deal of the AI conversation, when it gets sidetracked, reminds me of the old musician's union nonsense of confiscating Mellotrons and attempting to demonize Dr. Moog. So we need to keep the conversation leveled at the theft of intellectual and physical/audible property and stay away from demonizing the programmers as anything other than thieves.
Excellent and balanced points - and it’s obviously our duty (as digital luthiers) to
embrace new technologies that can advance our field. I do believe convincing AI instrument technology has to built ground up - and not top down as current. Has to be built by people that really understands the nuance of instruments and performance - otherwise just pale, emotion-deaf, cold clones.
I hear you on the paranoid attempts to stop technologies in the past. But see this more in the camp of Napster then Mellotron/Moog.
The tech is here to stay. It is just a matter of how to source responsibly.
The problem being that if AI companies train without consent. There won’t be any creative companies left to generate new “data” - and then suddenly progress would halt and stop at the level it was trained at. I don’t think that’s gonna happen, but it is the dark side of the equation.
It amazes me how little we value of our human qualities in a way that we, so quickly, become afraid of some technology kicking our creativity to the curb. I think we have to re-evaluate what the word 'intelligence' really means for us and how we define ourselves, and then AI will not be seen as such a threat. AI in music is really just another step up from the Midi Chord Pack (poor man's AI) , but it will stay and will definitely change the business of music, and the only 'real threat' it poses is that things will not be continuing the way we all know and have grown comfortable with...
Agreed. It is gonna get there at some point - and in many different iterations we can’t begin fathom. But as of right now - it’s still a plasticky and just making what’s already bad - even worse or more cliche. I suspect in part because some of the people building it haven’t spent their entire lives in the field. It’s a lot more complex when it comes to true human expression, which is the holy grail of sampling - with or without AI.
My belief of where the AI data comes from is the DAWs we use.
Every DAW ask if you will allow them to collect data in the background.
They say it’s to help better understand how we use and to help make the DAW better.
They say it’s collected anonymously. Who’s to say it’s not anonymously added to the AI database.
Just my thoughts.
It's more like Ai is making derivative works the same way we do after consuming years of music and then regurgitating something "original". All art is build from the process of accretion, I feel like AI works fit into that dynamic as well.
I think my fear of AI taking over the arts isn’t about it creating something better or more innovative than a human can, its about some company flooding every artistic space/platform (iTunes, Spotify, instagram, UA-cam, etc….) with substandard AI mediocrity thus drowning us lowly humans. The profit and takeover model is the issue. Great art won’t ever get seen or heard because it’s drowning in a sea of AI.
Right. We see it already. Try to do an image search on any stock image site - and it’s already populated with AI 💩. But in a way it was already that way with cheesy stock pictures. Fake smiles and someone holding a violin the wrong way. In a way … all this noise with just pollute the services more - leaving room for more curated services with high grade components. Some want McDonalds … others a non-machine brewed cappuccino.
We need more companies focusing on AI tools can help musicians create music and less companies creating AI's that make all of the music.
Hejsa Troels, sådan mand! :)
Ah, 1000 tak ven, kh T
Pienso que en éstos primeros intentos artísticos con la AI, es algo absolutamente sin vida. Puede funcionar en otros ámbitos de diferentes disciplinas de la vida. Pero en el hecho artístico, todavía no.
Yup. Unless “Soulless” by Manuel means becomes the new thing … wouldn’t bet my ears on it. 🥹
This is an interesting question(s) to ask is I don't know if there truly has an answer. There was a similar topic posted by Venus Theory within the last month, and he took a different take than you have had. But it covers so much in the grey area of music that I don't know if you can get a definitive answer.
When you are defining even the side of A.I. in music, there are some great tools out there that can be used to generate a set of midi notes that say it is A.I. Now without seeing the actual code that is used to generate those files, it could be just a large algorithm, but equally could be using some generative A.I. model to help in the process. But this brings up some pretty valid questions that I don't know if you could answer. Does the simple act of having a computer generated set of notes constitute that its no longer the work of an individual artist? You need to consider that they are just a set of notes of varying tone, duration and intensity strung together. Now if you simply selected import on a DAW and then took the default assigned instruments and published it, then I might agree with you. But what happens when you start to change things? Does it ever become the artists work? At what point? or is it forever deemed an A.I. computer produced piece?
So for example, if you go and change the assigned instruments that in onto itself can change the sound and feel of any musical piece. How about taking sections of the notes and changing the octave they are played in. Or even taking a progression of notes and changing their key. How about if you add / subtract notes from the piece? Take the order of when certain notes are played and changing the order in the score. By this time its getting into the realm of getting inspiration from some other music composition and making your own piece, but then it can be asked how much must you change in a whole song to make it all your own? I am sure that you can listen to music and pick out sections of any song that are very similar if not identical to other music that you'd call a totally different song even with the similarities.
I don't think that A.I. questions in music can be captured by a simple cut and dry model because there are too many examples in music history that will break the rules that are set and so the only answer can be a case by case model.
Yeah - and how to truly express. The rawness of the human touch. No two notes never truly being the same. And to your point the importance of breaking the rules, but in the right way. A sense of what is beautiful. Cheers, T
@@Soundpaintmusic thank you for responding, and for the wonderful soundpaint application. I have used it in several compositions and it is wonderful to work with.
I have used A.I. as a tool for equal access, to overcome my disabilities and allow me to access music and creativity that would otherwise be impossible. I was in a catastrophic automobile accident that left me with a traumatic brain injury. I am using music as a tool to help my brain recover the neural pathways that were damaged by the accident. I know I will likely never be able to learn how to play a piano or any musical instrument as a result of this accident, but determination isn’t going to stop me from playing.
I can use A.I. as a starting basis for laying down a rough template for a song or musical piece. And then note by note, chord by chord, edit or transform that is inside of the DAW to match what inside my head I can hear or imagine the piece to be. Assign the right instruments and through many iterations and variations develop the sound that I am looking for.
I’d like to think of myself as an artist and composer. There is the spark of creation inside of me that I feel raises me to that level. Who knows, as I recover more each year since the accident, perhaps I will get better to be able to write from scratch more songs, rather than using the current template model I am using. Each new piece I am learning more and more as my abilities grow.
Perhaps AI will be developed to the point where the human 'imperfection' component is utilized to the most perfect degree, unimaginable....
Perhaps. We don't know yet. (Im)perfection is a vast subject. One could argue "accidents", experiments and randomness too. Some of our greatest inventions came through it. Either its gonna require deep, deep and arguably multi-modal training in multiple and narrow subject fields to get even close.
That's a good point about the 'accidental' nature of some of our greatest human triumphs, possibly part of that imperfection aspect that is part of the intrinsic design of humans. The fact that AI came about because of us gives me some kind of hope that a breakthrough will occur within that arena; at least eventually. I'm not really sure if that 'randomness' of thought that is so characteristic of our beings can be engineered into a digital or otherwise format; but it may be our only hope of ultimately saving ourselves and our humanity from our own technological manifestations, which would possibly be the most ultimate irony imaginable....
True - though I doubt we would discover atoms, microwaves, penicillin through AI as it currently stands. It requires the ability to consciously observe - not just within a given set of parameters, but in fact outside of them like it happened with the above discoveries.
Perhaps 'consciousness' is the 'missing' component; an ability to 'think' both logically and in sequence, as well as corroborate external data, and then translate it towards a previously unimagined conclusion. There is also an element of transmutation involved; which AI seems to handle with a certain aplomb, although I'm not sure there is any real creative force at work in the traditional sense, but the results are often startling and seemingly unique in their hyperrealism. Whether AI can make any 'improvements' on an equation like E=mc2 is difficult to ascertain; but it still feels like very early days, and look where we've come with computers in a relatively very short period of time.
AI will take over. Watch Terminator 2. Judgement Day 2029!
It already has. It's already changing the news we read, the aesthetic choices we make (look at the mindless flood of a.i. generated imagery), the way we see artists and the act of creativity. We're already absorbing the poison and we're being put to sleep.
@@wietzejohanneskrikke1910 The AI generated imagery is just Amazing, soon AI will be doing the same for music. Just a matter of time, Judgement Day.
Or … the exact opposite … the technology eventually becoming a salvation to global warming … diseases … expanding human consciousness and empathy …
@@Soundpaintmusic There is no GloBull warming. Produce a chart for one location showing half a degree warming?
Data sets should be transparent. I think AI should be open source in general. If they want to stay closed source, they should have a 95% tax rate. But also companies that use open source should be taxed high, just less.
Progressive! I like it! Cheers T
AI wont steal my job. A creative person using AI will steal my job.
As for sampling source material and "training" on data, every musician I know who was taught music was trained on copyright material, and in most cases, they paid nothing to the original artists for training on that material (except for an attempt to do so here in Japan). Most professional projects I and colleagues have been a part of include a request to make something similar to copyright material, ideally in a way that is not so close that it could be considered infringement. I think we give professional music too much credit because its business via art, with more focus on the business than the art. AI is a tool that talented people will use well, and less talented people will use to just try and keep up.
One creative person with AI will steal 100 jobs , it can't be 1 to 1 . AI will learn from talented people to allow untalented people to do the same thing with no effort
Neutone Ai plugin has a "bulgarian Choir" sound.. we all know there's only one Bulgarian Choir company out there. I don't like that this is happening. It didn't sound good at all but like you say down the road a few years they might be able to perfect Ai. and this is why we need laws in place. Dreamtonics is all made with real human vocals , did you know this? like Solaria vocal bank is made by Emma Rowley, she's on youtube. It's similar to sample libraries in a way, it uses algorithms to work with the samples. very very advanced technology, i bet you would love to just talk with their dev for an afternoon, he seems like a nice guy, they are constantly making improvements to the software.
Wake me up as soon as AI makes me a variation of the Star Wars opening with the same quality. Until then, I don't care. I can make dull loops or simple chord schemes if I want, don't need an AI for that. I'm a writer for a music store (web content), and some of us are using chatgpt for their texts. Often you get the same dull results though, with countless variations of adjectives for the same subjects. If you want to iron all that stuff out, you'll soon find out it'd be faster to just write the whole text yourself.
AI is nice for images, I give you that, and I use it for fun. Wouldn't call it "intelligence" though, just a smart way of stitching existing material together. I mean, I tried an experiment once: "Bert from Sesamestreet holding a pineapple." Guess what... Bert was holding another Bert-head. That's not intelligence, that's just stitching matching shapes of given subjects.
Back to AI in music: the real problem arises when the audience couldn't care less about whether something is real or AI. For the simple reason they're bloody tonedeaf. Even worse: following audience who don't care, directors and executive producers for film/tv might choose AI as a way to save a buck and claim any music performance rights for themself.
Am I against AI in music? Not so much, as long as it's reserved for tools. I like tools, in fact it's an area which interests me, and I experiment with my own tools to create e.g. larger musical form structures, or suggesting chord progressions to go from situation A to situation B. Not because I couldn't myself, but sometimes it pays to have a weird 'third party' suggest something silly that was not on my initial radar. In Bob Ross' world everything was a happy accident. If composers become too good to make mistakes, who's going to make the happy accidents? Let AI suggest those instead.
As long as AI has to be told by a human what to create, and it is not an autonomous system, capable of deciding what it wants to do, it is not worthy of giving over our talent to it.
It has always been a case of "talent not included" when it comes to powerful software and hi-tech musical hardware. AI is simply a way to let the non talented among us create something. Not wrong... not right. But it needs to be controlled and regulated. There needs to be some sort of watermark within the "creation" of an AI system.
Just like the "tech support" people you get on the phone are tied to a closed system database, so is the AI system. It's sample sample sample, rearrange, and regurgitate. Not created, just strained through a sock into a tin can!
First off, it may be a bit of an uphill fight to pass laws that ban AI from being trained on copywritten material in that we all learn and imitate our favorite artists so the AI in that regard is not all that different.
But assuming AI is not going away (rather it is just going to get better and better) and any such laws regulating its use will likely be a long drawn out affair maybe the best approach is if one can't beat it, then join it!
Thus one can use it as a tool, just like any other tool such as Kontakt, Soundpaint, Band in a Box and such that can enable one to create things that otherwise would be hard if not practically impossible to do with ones own limitations of resources and skills.
For example I just did that on my most recent UA-cam video I just some minutes ago uploaded to my UA-cam channel that used a number of high tech tools such as Sono, Audimee, Mixcraft 10, and yes, Soundpaint as well.
And making it was a lot more work than just clicking buttons though my roles in using theses tools was not so much as a performer or even music creator or tune writer but more of as a director and producer. But even so, there was plenty of room for creativity but it was fun too in that the AI offloaded a lot of the work that otherwise I would had to devote to creating it!
So the trick might not be so much to use AI or not to use AI but rather how best might one use it.
So that is my two cents on this topic.
The genie is out of the bottle. No debating that. It’s perhaps more what we ask it for. I would like to ask it to be our artistic co-pilot. To learn and adapt to our individual preferences as artists, so it can augment my ideas without putting its own plastic vanilla on all of it.
@@Soundpaintmusic
Agreed! For example I wish Sono would have the option to output MIDI files and chords symbols so that one could add parts to it and also that it had teh option to break out each MP3 voice as a separate track.
Also, it would be good to be able to have oy edit those tracks without generated anew piece but have it evolve the other one such as say one can tell it to add a 5 measure bridge in measure 10.
So one might have Soro for the casual user vs a Soro for the music content creator that has other tools (like SoundPaint) that could be used as well.
But one other (big thing for me) is I hate click tracks (with a passion in that the music needs to be expressive in tempo as well as pitch and timbre and such) and wish there was a good AI tool that could take a MP3 and detect changes in the beat and then add measure markings which then could be edited.
Band in the Box sort of can do this but is really hard to use and takes a lot of time, where an AI that could get to mostly right and easy to then edit is what would be great.
But the big money is in making the AI for everyone and not just for music creators so I am not holding my breath on these to happen anytime soon!
Another AI tool I would like to see is one that might find and suggest fixes to "wrong notes" where it would ask the user, did you mean that to be a G and not a G# and then one can sayes or no verses having to sift through ones own playing to find those muffed notes.
And even better if that AI then could learn one's style and better do the above based on how one plays.
Things like this are (in part) the real potential of AI in music.
Also an AI assistant that can better do things like key switching and the like so as to take more off the load of the performer would be great as well so that the user can concentrate on making music and not controlling MIDI options and the like.
I might give an example that is pertinent to you guys.
I have a lot of your string libraries and such but hardly ever use all of the settings in that it requires to much fiddling with key settings and adding more tracks.
But if you had an AI that could do that for me where I could just with text say make this staccato and the AI would figure out how to do that but also do it off line so that instead of having say three or four instances of SoundPaint or Kontakt open for that part the AI could fuse all that into just on library on the fly that just has those sound files so that it takes little memory would be super!
That could be a huge time saver as well as memory and through put saver in that the AI could do this offline and then create a custom instrument that has just what I wanted for the part and nothing more.
Also if I wanted to add (or remove) a different sound file the AI could easily do that without a whole lot of fuss.
So I am hoping you guys will hire someone that can use AI to do things like that in that those business that can adapt are the ones likely not to become the way of teh dinosaur as the world continues to change.
To further elaborate on this idea for a human musican say a trumpet player one need only write in text on the score Harmon Mute and the tRumpet player knows what to do. I wish that in a track I could just do te same and add a text tag that said harmon mute and teh AI would figure out how to find and plice in that sound file into a custom one so I would beevr have to mess with key swicthing or searching for MIDI files or having to open up a track just for the Harmon mute sound but have the AI on the fly do all that for me so that say at the end I have a track with a sustained trumpet, a Harmon mute, and a staccato trumpet as one (on the fly created instrument) with nothing else (e.g. I don't need any trills or swells in that piece or straight mutes and such) so that custom instrument is lean and mean and easy and quick to create.
Thus just as one would just have one trumpet player that can play all those then one can just have one track that can do so likewise.
Thus a track is like a player and not so much a sound. One might even have that player be able to pick up a sax and the put it down and pick up a clarinet just as a real sax player could do all, in jus one (on the fly custom AI generated instrument) instrument.
And even better if the AI could produce a video of that sax player playing his/her (that choice also under my control as an option plus attire options as well as move options) part with correct figuring, body movement, and that can pick up that sax or clarinet and then be added to band video of the piece.
Now that would be cool AI! And it would give me (much) more creative power not less!
One more thought on this. If you at 8DIO don't have the funds to hire an AI expert maybe you can team with some other company like PG Music to cost share hiring an AI expert and thus form a consortium where you then can share teh cost.
For it seems a great business model is create consortium to create and develop a new technology, then once that is developed one can each then fight one another for market share using that new technology. That way all can win... so long as no one gets to greedy!
This might sound like a bit of a stretch at first...but I think a.i. could be useful in the Permaculture technique called 'Companion Planting' - but for humans...so people can end up living in a place where they are inspired musically instead of in a place where they get flooded with cortisol on a daily basis from abrasive and toxic input.
I am not famliar with the concept of companion planting - will read up upon it! Cheers T
You know, I am always fascinated by your judicious and delicious questions. So much for reflection, I think that AI does not exist, it is simply a sum of dominant information which could very quickly lead us to an absurd compromise of repetitiveness and pull contemporary music even lower than it is today . Silence is the keystone of music, dynamics is its reflection. AI is servitude, it is made to trample human intellect and creativity. Then comes the interpretation (an orchestra or a modern score) if it is poorly conducted it can quickly become boring. To converse about the quintessence of music, I seek above all the PLEASURE, of listening to and discovering the music of other composers but also of being the one who will draw the sketch, this can perhaps be triggered by a smell, a light, an image, a sound, a memory...... and I will start to build an emotion. Thank you Troels, I am happy that you take the time to exchange your many thoughts, this is perhaps why your tools are designed differently and of such high quality. Have a nice week end. fabien
Thx Fab! So agree with your statements on silence. Silence and raw expression. Another thing about AI is that it’s mainly good at what’s popular - because it’s more frequent in the datasets it trains upon.
I wrote a little poem about it for my friends:
AIs chase what crowds adore,
Sunsets, smiles, botox, evermore.
Fast food logic, designed to appease,
Soul's depth frozen, in silent freeze.
In the world of AI, perfection is the law,
Yet it's imperfection that gives art its awe.
@@Soundpaintmusic ♥
Ok but what would be your take on already working ai-supported instruments like with synthV? You mentioned that it would be a lot of cpu usage, and while I can’t deny that using synthv can be a bit consuming if you just leave it for too long. With the program you can basically make super realistic vocals if you know what you’re doing and based on the voice model you’re using, I honestly really really think its a waste that this sort of technology hasn’t yet been applied to other fields like orchestral instruments or even classical singing itself (since synthv takes a more pop approach).
It’s a subjective discussion. I have yet to hear an AI that really sings its soul out and wails like only a biological creature can. It’s not to say it can’t be done, but that last and crucial layer of expression is still missing. Great for vanilla, but still missing the edge of singing. The true, vulnerable, performance.
SynthV is at least trained by a vocalist who's consented to their voice being used for that purpose, and who is getting paid for their voice to be used, so there aren't really any of the ethical issues of training an AI on other people's music (whether it will ever be able to rival the expressiveness of a real vocalist is another question though).
@@sburton84 yeah exactly, and thats where a lot of my thoughts on it kinda fall in because usually as far as your average ai dataset its usually either taken from the internet, or in the case of voices, out of vocal samples of the target individual. But for both of these cases the quality that you’re gonna end up with is gonna be limited to the reach and range of where the data is coming from.
Which is why I said that this sort of thing is a miss so far in terms of other fields of virtual instruments like with orchestral instruments or more classical vocals, if somebody was to get the musicians to agree to have their playing likeness be used for some sort of ai software, in the same way synthv gets their vocalists to have their banks and etc. And especially as far as the entire sample library industry is concerned, as I’m sure with their input some very very expressive results can come out. After all generative models can only do as much as what its fed, and so if its fed some real high end stuff… well I don’t doubt itll be able to do just that given enough data for sure.
@@Soundpaintmusicfor the last 20 years most vocals have been pitch corrected and vunerable performances have been edited out . In a way a lot of recent pop music already sounds like AI
@@Dystopian84 Mostly because the world is flooded by people who can't sing and need to be corrected. 🤣 And that may have been the choice made by executives who prefer a supermodel singing over a musician singing. And that may be because video killed the radio star.
There is no "right or wrong" with any of this, it's a matter of targeting what type of content and audience you're aiming at, connected to that "genre". Most of the "usual suspect" commercial content is already so heavily engineered, AI is just another layer added to the tool system. Ever notice how so much of the commercial content sounds the same . . . gee, I wonder why.
However, at the other end of the spectrum, some content really is unique enough to be its own genre (I sort of fit into that realm a bit, with synth instrumentals). I don't know if I'll ever use AI to spew out the next pop tune or whatever, but I do use all sorts of unique effects and modular synth engines to spawn elements that go into my creations . . . maybe it's a matter of co-evolution with AI as collaborative lifeforms. OK, it's a concept . . .
Music is emotional communication: interpersonal exchange by means of emotions. AI has no emotions and cannot feel the emotions of humans. Even if AI could 'understand' human emotions -- that is, identify, classify and imitate them in some way -- it cannot empathise, i.e. emotionally resonate, with them. Therefore, the notion than a musical artist would need AI to compose a musical work is absurd.
There are claims that some AI systems are on the verge of becoming sentient and self-aware. If that is true, I look forward to AI becoming intelligent enough to realise that it is mindless, soulless, heartless, unloving and unlovable and therefore deciding to turn itself off: AI suicide. Bring it on!
I doubt A.I will pump out any music like Frank Zappa and Allan Holdsworth anytime soon.
It can pump out Zappa right now, however only because it was trained on Zappa. It won’t be as good as Zappa, but it will sound like Zappa. It won’t be as original as Zappa, because it’s a mere low resolution clone.
The question becoming is it okay it trained on Zappa without permission?
What’s the difference between a machine training on the data and a human training on existing music?
Two things:
1. AIs can train far vaster and deeper then any human ever could. The current generations of LLMs have a base IQ around 160. Imagine that 10 years from now in terms of potency.
2. AIs are trained on data (for example music and samples) without permission, which is illegal. One can argue the AI is just training and is inspired like humans, but that won't hold if it goes to court. AIs don't have rights, nor are the laws around them very developed. So the difference is that its the human using them that is potentially at liability. Just like a self-driving car may cause an accident, but its still the person operating that is held liable.
If you wanna go deeper - here is ChatGPTs answer to your question:
1. Learning Process:
Machine Learning: A machine learning algorithm, particularly in the context of music, learns by processing large amounts of data. It identifies patterns, structures, and relationships within the data (such as rhythms, harmonies, and melody structures in music) without any understanding of these elements. The learning is quantitative and based on statistical models.
Human Learning: Humans learn music through a combination of listening, practice, theory, and often emotional engagement. Human musicians interpret music, understand its emotional context, and draw on personal experiences and cultural background, which influences their musical style and understanding.
2. Interpretation and Adaptation:
Machine Learning: Machines apply learned models strictly as per the statistical significance and patterns found in the training data. They can adapt to new inputs only within the scope of their training and lack the ability to interpret music in a humanistic or cultural context.
Human Learning: Humans can interpret and adapt music dynamically. They can play a piece differently each time, infusing personal emotion or responding to an audience. They understand and express the cultural and emotional weight of music, adapting their performance to the context or their mood.
3. Creativity and Innovation:
Machine Learning: Creativity in machines is based on the variety and complexity of the data they have trained on. They can generate new compositions by mimicking or recombining elements from their training data but do not "create" in the human sense. They lack intent or personal expression.
Human Learning: Human creativity involves inventing or discovering music in ways that can express novel ideas, emotions, or cultural statements. Musicians can break rules, innovate styles, and even create entirely new genres based on personal insights and societal influences.
4. Emotional Connection:
Machine Learning: Machines do not feel emotions; they do not experience the joy of a cheerful melody or the sadness of a minor key. Their "appreciation" of music is purely based on data patterns and probabilities.
Human Learning: Emotions play a crucial role in how humans experience and create music. Musicians often convey and evoke specific emotions through their performances, connecting deeply with their audience.
5. Purpose and Goals:
Machine Learning: The aim here is often about automation, aiding human capabilities, or achieving scalability in analyzing or generating music. It's utilitarian and focused on efficiency or novelty.
Human Learning: For humans, learning music can be for personal enjoyment, artistic expression, cultural preservation, or professional achievement.
Troels, aren’t you one of the first who took ‘life out of music’ by making samples from real human ? Should we not all play the drums, bass or hire an orchestra to play really unique music and sound ? Samples still are just a snapshot from what a human has to offer. I think, it will take a while but it will come, AI will give us the tools to really express ourselves 100 % connecting to our voice or brain. Just wait and you will see that you are a dinosaur in music evolution.
If so … lemme be a chicken … one of the few creatures still roaming since the dinosaurs. I agree it’s gonna get there. But what’s interesting to me is the process of how we get there - and ensuring the process doesn’t completely erode an industry’ already in transformation. Do it the right way.
Soon AI won’t need humans anyway. I get more help from AI than from humans.
Have you heard of UDIO??
Yes - gonna do a video on it
When I record my voice, it becomes a sample.
Get over it.
And when someone samples your voice recording … it becomes a sample too …
My AI songs are not just push a button and done lol. I don't think my AI Music sounds like anyone else's. its a new barley understood instrument that has its good and bad. To say that these models are trained on other people's data is strange. When we first learned to talk, our speech was trained a lot from other people's speech. Dunno, It's Complicated haha - Party on Garth!
Right. One of the most common arguments for keeping training data secret - is that’s it’s not sampling, but merely learning like a human.
Well … An AI is not a human, nor does it have rights yet.
Now that we’ve made that distinction we have to realize that anyone creating AIs are still under the law.
Completely practically speaking there is no. AI music generator without music to train on or sample from. We can debate the semantics, but ultimately it didn’t learn out of the blue and source is protected.
You have to ask for permission to train.
Granted the AI companies don’t wanna do that - because they become at odds with the law. But that’s their problem.
Ultimately. An AI cannot produce music without having trained/sampled on others music. Whether we call it training or sampling doesn’t matter.
One speculative solution is to open it all up. So people can sample and train from anything they want in open ways. Behind it is a world wide AI system that automatically distribute royalties and proceeds to those songs being used. Regardless of what we call it. Training or sampling.
The AI doesn’t sample from the copyrighted material, there is no massive database that it’s searching. It’s learning concepts. It’s no more copyright infringement than you are listening to something and being inspired by it and doing a “sound-alike”. You can argue it’s SIMILAR, but you can’t argue it’s the same. For it to be copyright infringement it’s always been using the actual material in the output. It’s never been “someone’s been inspired by it”. The question is always more … “is this the real recording? Is this the same notes?”
Image AI’s already ruled it’s not copyright infringement. That sets a precedent in music, since it’s similar copyright law.
Why would they ever rule in favour of it being infringment when it would mean all the big AI companies would be in breach of copyright? How do you think they trained GPT-4 even? All of the text on the internet! That’s not their data either! They should get permission by that logic for all text models as well.
I also think you’re wrong about Suno. You haven’t explored it enough, there’s way more than crappy EDM. “Mindless beat generators” Denial isn’t going to save us, or you’ll be like those artists making fun of Midjourney Version 3 … well they’re not laughing now are they? Now they’re angry. They’re angry because it’s too good.
That's just not true. You clearly don't understand how LLM's are trained.
Ed. Suno didn’t just learn what music was out of thin air. It is clearly trained on existing music. Whether we call it “training” or “sampling” won’t matter.
There would be no music coming out of it - without a musical source to train/sample from. We can debate semantics, but ultimately it requires music to train and that source has to be disclosed.
It’s gonna come down to the courts and hopefully lawmakers too. There are many solutions that could work - but let’s do it the right way and not have another Napster at our hands.
I am rebel by heart too, but even by rebel standards this is an erosive mentality to blindly just sample other people’s music and then package it as “training”.
@@SoundpaintmusicBut Troels, it is significantly different to using copyrighted content the way things have done up till now. Even if you don’t agree, it’s also legally an absolute lost cause (see below).
Try to put your biases aside man, it doesn’t matter what we want. You were really into Midjourney, weren’t you? Didn’t you see what happened with Artists? They tried to argue the same exact thing as you did, and they LOST. Now you could say there’s more chances in the courts for them to rule differently, but if you look at why they’ve lost you’ll see they won’t. You’ll also see that Open Source makes the courts ruling absolutely irrelevant.
There’s no database of material the AI is searching and then splicing together. The training is literally teaching the model concepts. This is exactly the same way all the LLM’s work. They train the AI on all the stuff on the internet, everything you and I and everyone else has written.
The Ai learns very much like a person learns, and it doesn’t matter that it doesn’t really “feel” it. Everything is PATTERN, and the AI can learn what we consider meaningful but the patterns in its vast training data of everything humans feel to be meaningful.
Traditional legal determination is to look at the actual PIECE. Is it the same recording? Is it the same notes? The answer would be NO. If the AI was ruled copyright infringement it would have to use a TOTALLY DIFFERENT STANDARD. That being something like, “if it’s trained on copyrighted work, all output is infringement regardless of how similar it is”. How close, or not, it sounded to any composer or musician’s music would be irrelevant.
But I don’t need to convince you of any of this. The battle is totally unwinnable for so many reasons, and I’m not going to do them justice trying to list them here:.
Look at how much of an impossible battle this is…
1. The artists LOST the court cases against the Image AI’s.
2. The copyright law for imagery is essentially the same as for music, and the manor in which these AI models train (image/music/video/etc) is also exactly the same. So any court not finding in favor of your argument already establishes a precedent.
3. You’re going up against the BIGGEST COMPANIES IN THE WORLD. You think Open AI and Microsoft is really going to lose this fight?
4. EVEN IF ALL COURTS EVERYWHERE ruled that Midjourney/OpenAI/Facebook/Google couldn’t use unlicensed material or scraped material from the internet in their models, IT WOULDN’T MATTER! (And just imagine the consequences of that btw…)
5. You’d also need to court to say it’s illegal to allow any user to use a media reference, such as an image prompt or music reference.
6. THE FINAL NAIL IN THIS COFFIN:
Open Source like Stable Diffusion lets you use whatever training model you like. There’s custom models for Stabe Diffusion on everything yo can think of. You can make your own, on your own work, you can use your own photos to make AI photos using your own likeness.
Open Source AI Music models will be here soon, and AT BEST we might stop companies like Suno and Udio. Where we might take away everyone’s toys for several months.
Big deal!
As soon as an Open Source Udio comes out, and it’s even better, and you can train it on ANY MUSICAL MATERIAL YOU WANT, please tell me honestly what you think anyone is going to be able to do to stop it?
The absolute best you could hope for is that legally if you could prove someone used AI, they’d have to “substantively change it” just like one has to do with construction kit sample libraries. But so what? There’s plenty of legal argument that there’s a lot of substantive work that can go into crafting the finished AI music/images, and making a “construction kit sample library” acceptable doesn’t take much. The limiting factor for construction kit sample libraries is that there’s only so much material, with AI it’s effectively limitless .
@@SoundpaintmusicTL;DR:
- Artists lost in the courts,
- They’re up against the biggest corporations in the world with the best legal teams, and aren’t shy about lobbying lawmakers.
- Open Source makes all of that redundant.
- You’d have to ban AI completely.
AI has no Emotion, will never have, and will never able to pretend emotion. Everything i saw/hear from AI is absolutely dead and boring.
Udio did have a lot of emotion, and you’re in denial if you don’t think so. Denial won’t save you.
@@Edbrad It's about machines and computer programs. I think i will take the following question to the admission test for my students: What kind of drugs do I have to take to believe that a computer program can have or generate emotions? 😁