Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

MuseNet

openai.com

Post reply on HN
Print view View original

131–140 of 189 posts

Re: MuseNet

#131

Post by CuriouslyC »

leesec wrote: post #123
TaupeRanger wrote: post #118

Earlier quoted context omitted.

Go is a game with a strict ruleset. The comparison is not valid.

One could say music is also a game with a strict ruleset. Indeed having played the piano for 20+ years now, I often think of how no matter how good I am, I won't invent a new form of music or even style of music, I am limited to the 'rules' of my piano.

That doesn't surprise me, if all you do is follow the rules of music theory on a standard piano. If you creatively deviate from music theory and modify your instrument (or create a new instrument) you could easily come up with something new, the question is whether anyone besides you would enjoy it.

Re: MuseNet

#132

Post by p1esk »

modeless wrote: post #42

I think this is far, far beyond any algorithmic composition system ever made before. It displays an impressive understanding of the fundamentals of music theory. Most previous attempts at neural net composition restricted the training set to one style of music or even one composer, which is pretty silly if you understand how neural nets work. It was obvious to me if you used a very large network, chose the right inpu…

How do you compare them to Aiva [1]?

[1] https://aiva.ai/creations

Re: MuseNet

#133

Post by lawfulcactus »

lucidrains wrote: post #80
01100011 wrote: post #78

Earlier quoted context omitted.

It's probably only a matter of time before we have a GauGAN like interface for synthetic music creation...so you could say 'i want a sad song with a soft intro and a buildup of tension here with lyrics covering these emotions and things which lasts 7 minutes'. ML/DL is coming for a lot of the grunt work. It's coming for us as programmers as well. It's probably a few years away, but ML/DL

Given how easy it is to train a Transformer on any sequence data, and given how plentiful open source code is, I'd say "CodeNet" is probably less than a year away. OpenAI will probably do it first given they already have the setup.

Can you explain? I'm not an expert on ML by any stretch of the imagination, but you'd think with the sort of stringent logical coherence required to construct useful programs, it'd be a pretty subpar use case. Or do you mean smaller-scope tools to aid programming, like linters and autocompleters?

Re: MuseNet

#134

Post by ArtWomb »

Wondering if synthesizing a single instrument rather than an entire MIDI file might be better, as each note could be evaluated via finite differentials. Transformer would use as input a single note from say violin or guitar. And "translate" the pitch and envelope to the new sounds. The physical simulation of music, rather than notational logic. Problem of long range order in the composition still remains unsolved, unless the model accounts for the importance of transitions in music theory. In Neural Machine Translation, characters, words, even grammatical rules themselves can pre-define such an implicit structure.

Enjoyable to the trained mind to be sure. Especially as a moment in history. But the tunes themselves lack soul. And jarring for myself as the background music I had turned off to catch the tail end of MuseNet's performance, was of a particularly feverish level of human expressiveness ;)

Bad Brains - Live at the CBGB's 1982 (Full Concert)

https://www.youtube.com/watch?v=2pUlNfdnsAM

Re: MuseNet

#135

Post by smoe »

rco8786 wrote: post #12

This seems incredibly applicable to musical scores in movies. I can imagine a product where the editor/director/someone inputs a handful of variables (mood, genre, instruments, etc) and timing requests (crescendo beginning at 30s and ending at 75s, calm period from 90s - 120s, etc) and out comes a musical score for the movie that matches up with their scene editing.

I see the applicability more for video games. There are a bunch of problems that don't exist in other forms of music, e.g. when you want to make the music adapt to user actions.

"The Double-Edged Nature of Video Game Music" https://www.youtube.com/watch?v=HZB2_hlgKoA

Wouldn't solve all the problems, but I think there could be very interesting results if a composer provides skeletons of themes used, and "AI" adapts them to world and player state dynamically.

Re: MuseNet

#136

Post by 0815test »

p1esk wrote: post #132
modeless wrote: post #42

I think this is far, far beyond any algorithmic composition system ever made before. It displays an impressive understanding of the fundamentals of music theory. Most previous attempts at neural net composition restricted the training set to one style of music or even one composer, which is pretty silly if you understand how neural nets work. It was obvious to me if you used a very large network, chose the right inpu…

How do you compare them to Aiva [1]? [1] https://aiva.ai/creations

Aiva is way better than this honestly, and I mean the pure AI part of it, that (AIUI) only really generates monotimbral, piano-like music (the orchestration in the full pieces they release was - by their own admission - done by humans, last I looked into it). You can actually tell that Aiva involves real serendipitous ("out-of-sample") creativity of the sort that AIs (and good human composers!) are best at. (But the openly-available models I mentioned elsewhere in the thread are still a bit better, TBH. At least IMHO, and when it comes to rewarding active listening.)

Re: MuseNet

#137

Post by squarefoot »

fastball wrote: post #50
wilg wrote: post #19

Earlier quoted context omitted.

Maybe more interesting for video games, where you can have a realtime input of gameplay variables.

That could be very interesting. Every gameplay a different soundtrack. Sounds fun.

Even more interesting if it allows variations in the same soundtrack according to different dramatic moments in the same scenarios. Example: a man is walking down the street vs a man is walking alone in the night down the street vs a man is walking alone in the night down the street unaware of a bright red dot on his back. All these scenarios could use the same soundtrack, but require different dramatic levels. The game would send data to the music algorithms so that the soundtrack would reflect the right mood.

Re: MuseNet

#138

Post by daveheq »

I just read today a tweet by one of the React Rally spokespersons how Hacker News is a toxic forum... yet it's the least toxic most data-oriented forums I've used in years.

Re: MuseNet

#139

Post by mzs »

> RELENTLESS DOPPELGANGER \m/ \m/ \m/ \m/ \m/ \m/ \m/ \m/ \m/ \m/ \m/ \m/ \m/ \m/

https://youtu.be/CNNmBtNcccE

> Neural network generating technical death metal, via livestream 24/7 to infinity. Trained on Archspire with modified SampleRNN. Read more about our research into eliminating humans from metal: https://arxiv.org/abs/1811.06633

> More albums https://dadabots.bandcamp.com/

Re: MuseNet

#140

Post by pvorb »

daveheq wrote: post #138

I just read today a tweet by one of the React Rally spokespersons how Hacker News is a toxic forum... yet it's the least toxic most data-oriented forums I've used in years.

Sorry, but I can't take anyone seriously writing this on Twitter, the most toxic platform of them all.
Post reply on HN