Follow Slashdot stories on Twitter

 



Forgot your password?
typodupeerror
×
AI Open Source Facebook Slashdot.org

Meta Open Sources An AI-Powered Music Generator (techcrunch.com) 39

TechCrunch's Kyle Wiggers writes: Not to be outdone by Google, Meta has released its own AI-powered music generator -- and, unlike Google, open-sourced it. Called MusicGen, Meta's music-generating tool, a demo of which can be found here, can turn a text description (e.g. "An '80s driving pop song with heavy drums and synth pads in the background") into about 12 seconds of audio, give or take. MusicGen can optionally be "steered" with reference audio, like an existing song, in which case it'll try to follow both the description and melody.

Meta says that MusicGen was trained on 20,000 hours of music, including 10,000 "high-quality" licensed music tracks and 390,000 instrument-only tracks from ShutterStock and Pond5, a large stock media library. The company hasn't provided the code it used to train the model, but it has made available pre-trained models that anyone with the right hardware -- chiefly a GPU with around 16GB of memory -- can run.

So how does MusicGen perform? Well, I'd say -- though certainly not well enough to put human musicians out of a job. Its songs are reasonably melodic, at least for basic prompts like "ambient chiptunes music," and -- to my ears -- on par (if not slightly better) with the results from Google's AI music generator, MusicLM. But they won't win any awards.

This discussion has been archived. No new comments can be posted.

Meta Open Sources An AI-Powered Music Generator

Comments Filter:
  • First you bring music down to the level of generated noise to fill shopping spaces with, then you step in with the automation.

    No, this is not an art form.

    • by jmke ( 776334 )
      > No, this is not an art form.

      Nobody claims that either.
      it's just a tool, can used by everybody, and artists will actually create something artistic with it
      • by edis ( 266347 )

        > No, this is not an art form.

        Nobody claims that either.

        it's just a tool, can used by everybody, and artists will actually create something artistic with it

        You came to the base, with these:

        "nobody claims painting is an art form",
        "nobody claims theater is an art form",
        "nobody claims sculpture is an art form",
        "nobody claims dance is an art form",
        "nobody claims literature is an art form".

        Where I challenge you with the opposite: these and music themselves are assumed to represent an art form.

    • by vyvepe ( 809573 )
      I think this kind of AI can work quite well because precision and correctness is not important. Teach it on the best hits. Add some image recognition to detect how people react to the music. Provide this feedback back to the music generation.
      • by Rei ( 128717 )

        It requires a collaboration between humans and AI to make something good. Generate a ton of stuff, have a human select the best. Split each riff into separate instruments. Mix and remix. Add in vocals. Remaster. Etc.

      • think this kind of AI can work quite well

        Then you are sadly mistaken.

        I have tried it, and it sounds like a cracked 78RPM in a wind-up record player with a sock stuck in the horn.

        (Trained on "Cumbia Amazonica").

      • by edis ( 266347 )

        Dear folks, you are just guided away from what the music is. Yes, you have been dumped a lot on you, with a "music" moniker attached to it. Still, this process of sinking you in junk has never changed the idea of sensational power from creative sound.

        Thus. You can have variety generated. But you don't generate creativity. Here your AI will stumble and flap.

        This is where humane will emerge irreplaceable and spirited, unlike the machine, even though latter has helped a lot, and still would.

        However: music, by

    • by Anonymous Coward

      I shouldn't worry, I'm sure the RIAA and their ilk will soon be along to ensure the "quality" stays up.

    • Comment removed based on user account deletion
      • by edis ( 266347 )

        Just wait until one of these obviously automatically generated pieces of "music" turns into a song virus, or you find yourself singing along to one *shudder*

        Unless you are hopelessly processed former human being, this is not to happen to you.

        Rejoice!

  • Commodore 64 (Score:4, Informative)

    by Errol backfiring ( 1280012 ) on Wednesday June 14, 2023 @03:35AM (#63601204) Journal

    There was a music generator on the Commodore 64. Based on totally different algorithms off course, but I have a slight feeling time stands still when I read such news.

    You fed the C-64 program with a midi file, and you would have a few style options for the generation of the music. It generated variations on the input. Especially the "Fuga" option produced pleasant results.

    • Re:Commodore 64 (Score:4, Interesting)

      by mccalli ( 323026 ) on Wednesday June 14, 2023 @04:34AM (#63601270) Homepage
      Remember Band-in-a-Box? Yep. This is more sophisticated I bet, but this stuff has been around for quite a while in some form. It's very handy actually - I use it to generate things to just jam around with and while it doesn't make it into any final track I make, it certainly provides starting points to faff with.

      Logic's Drummer feature, as well as various other similar things, also fit into this category.
      • I spent hours playing with Band-in-a-Box. I'm not a musician (although I used to be an audio engineer), but it was lots of fun to play with!
      • by wfj2fd ( 4643467 )
        Algorithmic music generation isn't new. But the techniques used here *are* novel. The earliest approaches were generally based on things like hidden markov models, and later ones started to include other statistical techniques, but what they (BiaB and the like) were primarily note/instrument driven, they generated the MIDI control track which was fed to the synthesizer/instrument. These new models are directly generating the sound/waveform itself (sometimes through an intermediary like a spectrogram, but th
  • And you though music was already at the lowest common denominator.

    • by NFN_NLN ( 633283 )

      Wake me when AI can create absolute bangers like "Baby Shark". You just can't emulate that kind of talent with artificial intelligence.

  • Yet.

    But we're only talking about good musicians. All the others have been easily replaceable by machines for many years - or worse, machine-assisted to bring them to a barely tolerable level: quite frankly, replacing the current crop of "artists" relying on autotuners by good AI music would be a massive step up.

    • All the others have been easily replaceable by machines for many years - or worse, machine-assisted to bring them to a barely tolerable level:

      Agreed. Here's a quick video [youtube.com] on how this is done.

    • I've heard some "Drake AI" songs that were indistinguishable from the real Drake. When your vocal delivery is "impassive robot," and you're already outsourcing your writing, it's not hard to replicate by machine.

      It's got little to do with the autotune itself though. I've heard AI Paul McCartney that sounded indistinguishable from the real thing, vocally. ( https://www.youtube.com/watch?... [youtube.com] ) Though it was performing a song he actually wrote, so the tune and lyrics had nothing to do with AI.

      I would be intere

    • It's not even enough to put bad musicians out of a job.

      I asked for "up-tempo, a cappella doo wop with call-and-response style"... and it gave me some noise that was vaguely mariachi-like.

      So I fed it a clip from Dion and the Belmonts' "I Wonder Why" with the same prompt... and that was even worse, just random static noise with no tempo or melody.

      In other words, the quality is about on a par with the grotesque distortions that the art generator creates and the "alternative facts" that chatGPT hallucinates.

  • by The Evil Atheist ( 2484676 ) on Wednesday June 14, 2023 @05:23AM (#63601326)
    It removes duplicate parts from Cat Stevens songs.

    It's based on the principle Don't Repeat Yusuf.
  • These links point to a different company.
  • Itâ(TM)s kinda fun to play with and certainly very early days still. Seems good for âoecopyâ type of generation although it seems that stock sound sites are safe for now. https://youtube.com/shorts/MUU... [youtube.com]
  • 10 second Sound clip generated with this tech. https://youtube.com/shorts/_cn... [youtube.com]
  • Releasing the code begs cryptographers to advance with the automation at will, freely.

  • I agree with previous replies. The AI won't replace good musicians / composers. However, I believe the target is for background music for a video creator (like a Youtuber). For Linux folks interested in exploring the AI ... https://github.com/tkocou/Audi... [github.com]
  • At least for the prompts I was giving it, the first 3 seconds or so was promising and then it was 9 seconds of repeating the 4th second of output.

    Maybe they should have limited it to 3 second samples instead...

  • Musical creativity works by finding things that people like. If it can't predict or produce something NEW that people will like, all it can do is produce music that, oh that sounds like a crappy Beatles or Lady Gaga song, just not very good. It needs to be able not only to generate, but also to appreciate. Not impossible, but not yet achieved.

  • to write a song 'Donny 99'.
  • by MpVpRb ( 1423381 ) on Wednesday June 14, 2023 @11:19AM (#63602136)

    ...and neither is much of today's pop music.
    The way today's pop is created, it might as well be made by robots.
    The creators follow roughly the same algorithm as chatbots, making stuff that is strongly based on what is popular, with just the tiniest bit of difference so it isn't a direct copy.

    Meanwhile, creative artists play for free in their basements

  • On the one hand, I don't want to touch this, lest someone someday might say I didn't create the music I composed, recorded and published.

    On the other hand, my music likely was included in the AI training, so in effect I'd arguably be, at least partially, ripping off myself.

"The vast majority of successful major crimes against property are perpetrated by individuals abusing positions of trust." -- Lawrence Dalzell

Working...