Artificial Intelligence that makes dubstep

[Salfa]

@ Salfa

What AI tecniques will you be using, if you can share?

I debated which one to use. In the end, it was between doing artificial neural networks with biological algorithms, and statistical ai. I decided statistical ai would be the best, since the other option would take a long time to teach. When I say teach, I really mean it. When using biological algorithms, you actually have to teach the ai
What to do, you cannot program it in. unfortunately, statistical ai does not give any improvements to general ai like a.n.n. would. Statistical Is too specific for that. But that's why I chose it, it's very specific.

[Today]

so you still have to do all the sound design and sample selection, right? and mix procedures... this is just an arrangement bot?

[joegrizzly]

Fuck Robots, I seen that will smith movie

[ScarletCyanide]

As much as I somewhat agree in that it would be cool to do, I can't help but think it's a bad idea... I mean, do you really want your music made by a computer? If you do, do this, please never release the code... Make millions, become famous, never tell people how you did it... The last thing we want is for our last truly human resource to become mechanical, sterile and produced on an industrial scale...

[Today]

the programmer would still have to synthesize or otherwise design/select all their sound samples, and then tell it to place them in a melodic, harmonic, and rhythmic context that communicated tension, resolve, call and response, sequencing of motifs into a coherent passage, and finally arranged into a structured recording

and then you'd need to mix it, including all volume, pan and FX parameters, all their automation, down to the length of a reverb tail.
any bussing, sidechaining or other routing, etc etc

then master it

in short, it can't... excuse me, won't be done.
No realistic amount of coding will accomplish the above processes with anywhere near the competency that even an untalented human could.
Music composition, engineering, and production jobs are safe. This is just a roundabout way of generating a sequence, given inputs and assignments, and to be frank, other than setting a personal goal for the programmer themselves (which is not even really that ambitious, no disrespect), it is completely useless to humanity.

[corpu5]

jesus christ he's 17. I did programming at uni, got 40% in that module. well couldn't be arsed

[cosmic_surgeon]

Yeah if he keeps that up he'll be wedged when he's older. Very skilled programmers make so much money, and there seems to be a lot of high paying jobs for them atm.

[e-motion]

@ Salfa

What AI tecniques will you be using, if you can share?

I debated which one to use. In the end, it was between doing artificial neural networks with biological algorithms, and statistical ai. I decided statistical ai would be the best, since the other option would take a long time to teach. When I say teach, I really mean it. When using biological algorithms, you actually have to teach the ai
What to do, you cannot program it in. unfortunately, statistical ai does not give any improvements to general ai like a.n.n. would. Statistical Is too specific for that. But that's why I chose it, it's very specific.

Yeah, I know about neural networks and genetical algorithms (I had 2 subjects of artificial inteligence @ university, although it was more linked with the classic AI like Case Based Reasoning, Fuzzy Logic and some other that I can't explain in english but we basically used Prolog).

But yeah it's kind of a too complex problem to model...

[jrisreal]

While I don't agree with the concept, the project sounds interesting. I'd like to hear a tune made by it once its done.

[sunny_b_uk]

wow your 17 and programming, its so difficult!
i done a little DSP coding in a VSTi i made to shorten CPU useage and it was pure headache

[Salfa]

so you still have to do all the sound design and sample selection, right? and mix procedures... this is just an arrangement bot?

no it will do all of that too. i want it to be completely independent.

[Salfa]

the programmer would still have to synthesize or otherwise design/select all their sound samples, and then tell it to place them in a melodic, harmonic, and rhythmic context that communicated tension, resolve, call and response, sequencing of motifs into a coherent passage, and finally arranged into a structured recording

and then you'd need to mix it, including all volume, pan and FX parameters, all their automation, down to the length of a reverb tail.
any bussing, sidechaining or other routing, etc etc

then master it

in short, it can't... excuse me, won't be done.
No realistic amount of coding will accomplish the above processes with anywhere near the competency that even an untalented human could.
Music composition, engineering, and production jobs are safe. This is just a roundabout way of generating a sequence, given inputs and assignments, and to be frank, other than setting a personal goal for the programmer themselves (which is not even really that ambitious, no disrespect), it is completely useless to humanity.

Well i'm going to disagree with you. The program will do everything you just said. This isn't a stupid 3 week project. Projects this big take months of time. So far I'm probably only around 5% done with the whole thing, even less.

But I will agree that this is mostly useless to humanity. Don't forget I'm still a kid, I can do whatever the hell I want as I don't need to support myself.

[Today]

dammnnnnnn
i didn't realize it was going to do all that.
reinforced my premonition that you are in fact a megagenius

[blinx]

cool idea, wish you the best of luck....

what daw will you be using if/when you get to the point of running this for reals?

[Efrafa11]

http://www.link.cs.cmu.edu/melody-generator/

Its a melody generator... Could help in shed some light on what you could do for the music theory side of this idea.

[bassinine]

why not program the randomness into it?

i don't know if this has been mentioned (or even, if it will work) but..

why not do something to the effect of: 1 = 16 bars 2 = 24 bars 3 = 32 bars (then randomize it, giving a 1/3rd chance to select either 1 through 3. seems to be a way to inject variation.

[Salfa]

why not program the randomness into it?

i don't know if this has been mentioned (or even, if it will work) but..

why not do something to the effect of: 1 = 16 bars 2 = 24 bars 3 = 32 bars (then randomize it, giving a 1/3rd chance to select either 1 through 3. seems to be a way to inject variation.

I don't quite understand what you mean. I mean I understand what your saying, but what should I apply it to? And I am doing statistical AI, so it's pretty similar to what your suggesting, just maybe 24 bars has a bit higher chance. (Thats an example)


EVERYONE: Sorry I haven't been updating or replying recently, I have been sick, but I'm now feeling better and will continue production!

[Salfa]

http://www.link.cs.cmu.edu/melody-generator/

Its a melody generator... Could help in shed some light on what you could do for the music theory side of this idea.

Thanks! I added the link and your name under the credits of Music Theory!

[Efrafa11]

http://www.link.cs.cmu.edu/melody-generator/

Its a melody generator... Could help in shed some light on what you could do for the music theory side of this idea.

Thanks! I added the link and your name under the credits of Music Theory!

Just to be clear I didn't create that program but when I saw this thread it came to mind

[Salfa]

http://www.link.cs.cmu.edu/melody-generator/

Its a melody generator... Could help in shed some light on what you could do for the music theory side of this idea.

Thanks! I added the link and your name under the credits of Music Theory!

Just to be clear I didn't create that program but when I saw this thread it came to mind

I understand that, but I wouldn't have found it without you. So I credit you with the discovery of it. haha.