User Controls
I made a thing
-
2018-11-10 at 11:28 AM UTCIt plays piano:
Bretty slick huh?The following users say it would be alright if the author of this post didn't die in a fire! -
2018-11-10 at 1:27 PM UTCit would be a lot cooler if i knew what was going on
-
2018-11-10 at 4:12 PM UTC
-
2018-11-10 at 6:27 PM UTC
Originally posted by OG_GREENPLASTIC_JOHNSON_III it would be a lot cooler if i knew what was going on
It's a program for quantizing and building markov chains out of MIDI streams. So like basically you play a MIDI instrument to it, or in this case play it a MIDI recording of a performance, and it infers a probabilistic model of the piece which it uses to generate "new" music. You get a some influence over how faithful it is to what it was trained on, at the lowest level of fidelity it's basically just a random salad of notes that happened to appear in the training data, and at the highest it just plays back what it "heard", but there's an interesting middle ground where it takes little sequences and themes in the training set and splices them together that actually sounds intelligible, sometime it's surprising how coherent is, sometimes it gives an impression of actually having some kind of narrative structure even though it's just a statistical model and any larger kind of structure kinda just arises by chance.The following users say it would be alright if the author of this post didn't die in a fire! -
2018-11-10 at 6:41 PM UTC
Originally posted by Lanny It's a program for quantizing and building markov chains out of MIDI streams. So like basically you play a MIDI instrument to it, or in this case play it a MIDI recording of a performance, and it infers a probabilistic model of the piece which it uses to generate "new" music. You get a some influence over how faithful it is to what it was trained on, at the lowest level of fidelity it's basically just a random salad of notes that happened to appear in the training data, and at the highest it just plays back what it "heard", but there's an interesting middle ground where it takes little sequences and themes in the training set and splices them together that actually sounds intelligible, sometime it's surprising how coherent is, sometimes it gives an impression of actually having some kind of narrative structure even though it's just a statistical model and any larger kind of structure kinda just arises by chance.
translation: its a kike-esque copyright infringer. -
2018-11-10 at 6:50 PM UTCthat song reminds me of that time i went into that hyroglyphics store barnds and nobles but then i got bored cause who knows wtf that bound papyrus gobeldyguk sleeve of farts even means lol
fuckin eggheads in there just staring at them for hours like they got som kinda idea ( AYE CABRON, can somebody pass the grey poupon before i ASS?! )
.
( feds moved my dish soap while making this phred )
. -
2018-11-10 at 7:06 PM UTC
Originally posted by A College Professor that song reminds me of that time i went into that hyroglyphics store barnds and nobles but then i got bored cause who knows wtf that bound papyrus gobeldyguk sleeve of farts even means lol
fuckin eggheads in there just staring at them for hours like they got som kinda idea ( AYE CABRON, can somebody pass the grey poupon before i ASS?! )
( feds moved my dish soap while making this phred )
its in the same place as your soap-on-a-rope
.
tied around your ankles.
,
, -
2018-11-10 at 7:31 PM UTCthat was highly inappropriate
i think you need to watch the video again, dum dum -
2018-11-10 at 8:47 PM UTC
-
2018-11-10 at 9:15 PM UTCWhat is the name of this program and where can I get it?
-
2018-11-10 at 10:50 PM UTCThat is pretty dope, but can it help me be the next lil peep?
-
2018-11-11 at 6:05 AM UTC
Originally posted by AngryOnion What is the name of this program and where can I get it?
It's called Max and/or Max/MSP. It's a shit name and a pain in the ass to google for so you end up leaning on the built on documentation system a lot, which is fortunately really good. It's a bit pricey, I think it was like $80/year for a subscription or you could buy a lifetime license for a few hundred. There's a open source fork/re-implementation called "pure data" which is free, it's actually from the original author of Max but there's no development budget so it's a lot uglier and the documentation isn't as good, but it's fairly close to feature parity.
I'm really very critical of "visual programming languages" like this. I've personally had the pleasure of trying to use them in an educational environment and there's this stunningly consistent trend where there's some collection of examples that the authors thought about and which work well but then like the first idea a student has that's not in the tutorial set you need to break out some ungodly nested tangle of visual components because even simple things like "hold onto a value until I get some other value" are just fucking inexpressible because no one ever uses these shit toy languages to actually do anything, they just foist them on students. I was assistant teaching a course using Scratch (a visual programming language) like a year ago and a kid asked me how he'd do a fairly simple thing to what was genuinely creative, exploring the mechanics of what he'd been taught, like fucking exactly the kind of thing a bright student should be trying to do and I had to basically tell him sorry bud, exactly the right idea, you're way ahead of the class, but the amount of scaffolding and number shuffling and hacks that it's going to require as a result of the broken language educators inflicted on you is just too much for you to ever realistically be able to accomplish.
THAT SAID, I've actually grown to have a certain fondness for Max. Basically it does the right thing and kicks you in the nuts pretty quickly after you get started, but because it's willing to let you stumble a little bit on the learning curve what you end up with is a system that's actually reasonably usable. You'll curse the weird "things on the right executed before things on the left" thing and the hot/cold construct feels like it's a nuisance at first then you fuck around with it for long enough and realize there's a method to the madness and it ends up being a pretty usable system with some features that, unlike the vast majority of "visual" programming languages, actually gives you some advantage to the tried and true text file.
Originally posted by OG_GREENPLASTIC_JOHNSON_III That is pretty dope, but can it help me be the next lil peep?
Unless you think there's an untapped of market of young americans who are really in the mood for someone to fuse emo, rap, and algorithmic composition I'd say your odds are pretty low here. -
2018-11-11 at 6:13 AM UTCI was expecting your own nigga-speak code for an intelligent program of your own.
Fail -
2018-11-22 at 10:39 PM UTC
-
2018-12-02 at 12:59 AM UTC
-
2018-12-02 at 1:51 AM UTCI prefer to start go fund me's, it's easy to tug on soft hearts for a few bucks x 1000
-
2018-12-02 at 4:07 AM UTC
Originally posted by Esplender Hella cool. How does one start doing this sort of thing? I kinda want to dabble in the whole machine learning/AI thing but don't know where to begin.
By this sort of thing do you mean algorithmic composition/computer music type stuff, or ML/AI in general?
If you want to make music with computers, there's lots of resources for learning to use a DAW like logic or ableton or whatever else. I can't really recommend anything since I don't have much experience there. I've done more direct audio programming, which is kind of hard to get into. You end up needing to know how to program, then a bunch of stuff about digital audio, and then a lot of nuts and bolts stuff about scheduling and working with your OS's audio API, and then some music theory on top of all of that to actually make music. It's kind of a black art. Max/MSP (software in the video in OP) is kind of a middle ground between those two things, it's a lot more flexible than a DAW and you end up working with an actual numeric signal a lot but it ships with a lot of that nuts and bolts stuff that you can spend ages agonizing over in direct audio programming.
If you're interested in ML/AI I'd say go read a book or brush up on introductory stats and linear algebra, then read the landmark papers. There's an absolute deluge of garbage "learn AI from youtube/online learning thing/whatever in X minutes" style material on the internet and published as books and you want to stay as far away from that as possible.
