Ok the NI orchestral and voice samples are super impressive because I was under the impression acoustic samples sounded pretty flat still. It's kind of incredible the AI-generated stuff manages to sound so bad given the accessibility of great samples.
It seems like Magenta has been trying to sell the AI-tools-for-musicians angle since pre-LLM times, and I'm sure there are many others, but seems like they haven't really taken off? Why do you think this is?
FYI this was flagged as a promotion by my spam filter, likely due to
> Companies like Native Instruments have streamlined creatives’ access. They have built a future where anyone can make great music. Whether you're a shower singer or a charting artist, companies like Native Instruments have broken barriers between the musician and the songs they dream of making. No instruments needed, just imagination, a computer, and a checkbook. It’s from your mind to music.
I used to be one of those 2010 kids making wobbles and growls.
I met some of the producers that are now amongst the most influential producers in the dubstep scene and I'll tell you that making those sounds was THE challenge.
Frequency mod, ring mod, resampling you name it, we tried!
FOR SURE this will level the access but I am not so sure this is the optimum tho (maybe an RL will tell us where it is one day lol)
I am now a researcher but I used to love the sound design process, it was always unpredictable, and some of the best sounds just came out of experimentation.
About moving all this to AI?
Not even sure if this is event the right question to ask, but I hope the answer is NO and I am confident that people will keep experiment and 'generate' new sounds and not just emulate existing ones using these kind of tools.
I just find it depressing that these companies can raise millions of venture dollars for garbage and no good tools seem to be emerging. Like, can you imagine a diffusion based search of a preset library? That could be wild. What would your dream AI music app look like?
Dissuasion search of preset libraries would be amazing.But honestly I would be satisfied even with just an AI assistant that manages all the boring stuff for you like make automations in the DAW, opens new instances of VSTI, plugs effects without requiring you to do manually (just with some vocal commands). I think this would seriously improve the flow of the ideas: reducing time of execution from artist brain to DAW.
I never used FastTracker II because I had already moved on to Sonic Foundry Tools (Acid and Vegas). I have never seen Buzz before. that's pretty cool. In the 2000s, I used Fruity Loops for a bit, but was an Ableton Live beta tester and have been a loyal customer for 23 years.
I'm confused af by your post. Does anyone believe that the new AI music companies let you make sounds that are hard to make in other software? The selling point is (of course) "type in a text prompt, get a customized song out," all while basically knowing nothing about music. I believe the songs you linked were all constructed by people who know how music works.
The same is true for AI images. (Almost) any image made by Midjourney or whatever is pretty much makeable by a sufficiently trained human, it's just that the program lets you make an OK approximation with almost no money and no knowledge of how to make art yourself.
I fully agree that the songs being generated today are uncanny valley awful, and I also predict their eventual niche will be much smaller than for AI images. But I'm not sure what point you're trying to make here.
Oh, I'd much rather use everything you described than any text-to-finished-song garbage. The current apps are amazing and seem pretty sufficient. I might use a text-to-texture or text-to-patch app maybe?
But then again, I'm generally pretty crusty about tools that attempt to help create art without knowing anything about art. I do admittedly use AI image generators sometime, but I feel bad about it. My nieces made me read them a ChatGPT generated story but I hated it.
Text-to-patch, Rif! I think we're almost there. I know there's 0 dollars in these ideas, but I would love a text-to-serum-patch or a text-to-battery-kit...
Ok the NI orchestral and voice samples are super impressive because I was under the impression acoustic samples sounded pretty flat still. It's kind of incredible the AI-generated stuff manages to sound so bad given the accessibility of great samples.
It seems like Magenta has been trying to sell the AI-tools-for-musicians angle since pre-LLM times, and I'm sure there are many others, but seems like they haven't really taken off? Why do you think this is?
Well, the AI people would have to pay for samples, and that's against their ethos.
I like the Magenta team and have played with their plug-ins. I could never get sounds I liked out of them, but everyone's different.
This is a great question, Galen. A seed for the next blog.
FYI this was flagged as a promotion by my spam filter, likely due to
> Companies like Native Instruments have streamlined creatives’ access. They have built a future where anyone can make great music. Whether you're a shower singer or a charting artist, companies like Native Instruments have broken barriers between the musician and the songs they dream of making. No instruments needed, just imagination, a computer, and a checkbook. It’s from your mind to music.
As it should be!
Wow!
I used to be one of those 2010 kids making wobbles and growls.
I met some of the producers that are now amongst the most influential producers in the dubstep scene and I'll tell you that making those sounds was THE challenge.
Frequency mod, ring mod, resampling you name it, we tried!
FOR SURE this will level the access but I am not so sure this is the optimum tho (maybe an RL will tell us where it is one day lol)
I am now a researcher but I used to love the sound design process, it was always unpredictable, and some of the best sounds just came out of experimentation.
About moving all this to AI?
Not even sure if this is event the right question to ask, but I hope the answer is NO and I am confident that people will keep experiment and 'generate' new sounds and not just emulate existing ones using these kind of tools.
I just find it depressing that these companies can raise millions of venture dollars for garbage and no good tools seem to be emerging. Like, can you imagine a diffusion based search of a preset library? That could be wild. What would your dream AI music app look like?
Dissuasion search of preset libraries would be amazing.But honestly I would be satisfied even with just an AI assistant that manages all the boring stuff for you like make automations in the DAW, opens new instances of VSTI, plugs effects without requiring you to do manually (just with some vocal commands). I think this would seriously improve the flow of the ideas: reducing time of execution from artist brain to DAW.
Diffusion*
Did you ever use FastTracker II (late 1990's), Jeskola Buzz (early 2000's) or similar? Or did these programs never reach the new world?
I never used FastTracker II because I had already moved on to Sonic Foundry Tools (Acid and Vegas). I have never seen Buzz before. that's pretty cool. In the 2000s, I used Fruity Loops for a bit, but was an Ableton Live beta tester and have been a loyal customer for 23 years.
I'm confused af by your post. Does anyone believe that the new AI music companies let you make sounds that are hard to make in other software? The selling point is (of course) "type in a text prompt, get a customized song out," all while basically knowing nothing about music. I believe the songs you linked were all constructed by people who know how music works.
The same is true for AI images. (Almost) any image made by Midjourney or whatever is pretty much makeable by a sufficiently trained human, it's just that the program lets you make an OK approximation with almost no money and no knowledge of how to make art yourself.
I fully agree that the songs being generated today are uncanny valley awful, and I also predict their eventual niche will be much smaller than for AI images. But I'm not sure what point you're trying to make here.
You've worked in data-driven audio forever, what would your dream AI-enhanced music app look like?
Oh, I'd much rather use everything you described than any text-to-finished-song garbage. The current apps are amazing and seem pretty sufficient. I might use a text-to-texture or text-to-patch app maybe?
But then again, I'm generally pretty crusty about tools that attempt to help create art without knowing anything about art. I do admittedly use AI image generators sometime, but I feel bad about it. My nieces made me read them a ChatGPT generated story but I hated it.
Text-to-patch, Rif! I think we're almost there. I know there's 0 dollars in these ideas, but I would love a text-to-serum-patch or a text-to-battery-kit...