Want to read Slashdot from your mobile device? Point it at m.slashdot.org and keep reading!

 


Forgot your password?
Close
typodupeerror
×
Music AI

Mixed Reactions to New Nirvana Song Generated by Google's AI (engadget.com) 88

Posted by EditorDavid from the sounds-like-teen-spirit dept.
On the 27th anniversary of Kurt Cobain's death, Engadget reports: Were he still alive today, Nirvana frontman Kurt Cobain would be 52 years old. Every February 20th, on the day of his birthday, fans wonder what songs he would write if he hadn't died of suicide nearly 30 years ago. While we'll never know the answer to that question, an AI is attempting to fill the gap.

A mental health organization called Over the Bridge used Google's Magenta AI and a generic neural network to examine more than two dozen songs by Nirvana to create a 'new' track from the band. "Drowned in the Sun" opens with reverb-soaked plucking before turning into an assault of distorted power chords. "I don't care/I feel as one, drowned in the sun," Nirvana tribute band frontman Eric Hogan sings in the chorus. In execution, it sounds not all that dissimilar from "You Know You're Right," one of the last songs Nirvana recorded before Cobain's death in 1994.

Other than the voice of Hogan, everything you hear in the song was generated by the two AI programs Over the Bridge used. The organization first fed Magenta songs as MIDI files so that the software could learn the specific notes and harmonies that made the band's tunes so iconic. Humorously, Cobain's loose and aggressive guitar playing style gave Magenta some trouble, with the AI mostly outputting a wall of distortion instead of something akin to his signature melodies. "It was a lot of trial and error," Over the Bridge board member Sean O'Connor told Rolling Stone. Once they had some musical and lyrical samples, the creative team picked the best bits to record. Most of the instrumentation you hear are MIDI tracks with different effects layered on top.
Some thoughts from The Daily Dot: Rolling Stone also highlighted lyrics like, "The sun shines on you but I don't know how," and what is called "a surprisingly anthemic chorus" including the lines, "I don't care/I feel as one, drowned in the sun," remarking that they "bear evocative, Cobain-esque qualities...."

Neil Turkewitz went full Comic Book Guy, opining, "A perfect illustration of the injustice of developing AI through the ingestion of cultural works without the authorization of [its] creator, and how it forces creators to be indentured servants in the production of a future out of their control," adding, "That it's for a good cause is irrelevant."
This discussion has been archived. No new comments can be posted.

Mixed Reactions to New Nirvana Song Generated by Google's AI More

Mixed Reactions to New Nirvana Song Generated by Google's AI

Comments Filter:

  • Sounds like trash B tape (Score:5, Funny)

    by nadass ( 3963991 ) on Sunday April 04, 2021 @04:45PM (#61236410)
    I'll spare y'all the ear-sore of enduring the 3 minutes of badly-filtered, sound booth-isolating studio recording... ugh. It's just terrible. It SOUNDS like someone interlaced 5 distinctly-separate pre-fabricated badly-compressed MP3 files, then forgot to make sure they were in-tune with one another.

    By the end, it sounds like fan fiction by someone who doesn't understand how to make masterpieces. I was left with no connection with the material (i.e. the song) or the persons evoked within the rendition of the song (i.e. Nirvana or Kurt Cobain's digital AI).

    Of course, that's all just my opinion.
    • Consider that any lost songs discovered would be in a similar state: a work in progress demo recording. This "what if" should not be compared to a commercial release in terms of mastering quality, but if that's very important to you, agreed :).
    • My thoughts exactly. Kudos for beating the rest of us that enjoy real Nirvana music to commenting on this weird "song" that was made. At least we know that Skynet isn't ready for prime time.

    • Re: (Score:1)

      by amchugh ( 116330 )

      Generating something that sounds like a Nirvana track that got left on the cutting room floor is still an impressive achievement for an AI even if it would need work to even beat a track like Aero Zeppelin.

    • It all sounded in-tune to me. With MIDI in a DAW you need to go out of your way to make it sound otherwise. Go in and detune each individual note one-by-one, or make up a custom tuning file to use in your VST, assuming it even supports that. The characteristic underwatery sound of a poorly-made MP3 was nowhere to be found either... The only part that sounded off was toward the end, where the "guitarist" apparently lost his ability to keep time and ended up sounding like me. Probably they went overboard on t

    • Re: (Score:2)

      by AmiMoJo ( 196126 )

      That's the idea. Nirvana were a grunge band, their sound was lo-fi.

    • Well if the AI managed to write a Nirvana "hit" song instead of a b-side track things would get really complicated.

  • It doesn't matter if you don't like these specific songs, the underlying technology is very impressive. Forget emulating dead stars, I foresee this being used as a technology for generating new songs by using a collection of songs as a source.

    If you didn't think music was being manufactured before then this could change your mind as the process only requires someone to select the best generated content. I foresee a lot of music being generated by AI in the future which may or may not be a bad thing.

    • Re: (Score:2)

      by NFN_NLN ( 633283 )

      > It doesn't matter if you don't like these specific songs, the underlying technology is very impressive.

      This technology could put bands like Nickelback out of business. Their bread and butter is creating mechanical derivatives of their own material. With AI now filling that role, they are relegates to nothing but monkeys on instruments.

      • Their bread and butter is creating mechanical derivatives of their own material.

        You've basically just described the entire popular music industry. Thing is, rehashing the same old formula works. [youtube.com]

    • Musicians and actors should be very afraid of this. Today's pop music is almost entirely manufactured as it is. Eventually they won't even need an attractive person to auto-tune. And for existing 'musicians', don't like your new contract? Then GTFO and we'll keep making your music with AI.

      Macaulay Culkin made $100k for Home Alone 1. He made $4.5 million for Home Alone 2. Is that because he was suddenly a 45x better actor?

  • Sample lyrics are awfully close to "All Apologies" (Score:3, Insightful)

    by jwjr ( 56765 ) on Sunday April 04, 2021 @05:10PM (#61236450)

    "All Apologies", while not Nirvana's biggest hit on the charts, has over time become one of their best-known songs, and was the last song on their final album. It contains the lyric "In the sun, I feel as one".

    An extrapolation from this to "I don't care/I feel as one, drowned in the sun," seems pretty uncreative. A lyricist would probably avoid such obviously similar lyrics to avoid the perception of re-writing the same song. As such, at least the lyrics seem like a pretty unimpressive effort.

    • Re:Sample lyrics are awfully close to "All Apologi (Score:4, Insightful)

      by sound+vision ( 884283 ) on Sunday April 04, 2021 @06:51PM (#61236654) Journal

      I'm guessing the lyrics were generated with something like a Markov chain, which I've spent some time playing around with.

      You feed it a body of text and then it essentially spits it back to you in endlessly rearranged ways, but with an eye toward which words are associated with another. So small snippets (lines) can make some kind of sense when taken in isolation, but the more you read on, the more it becomes clear there's no overarching meaning or grand structure to the output. Using a Markov chain generator on IRC logs as a dataset, it's possible to pick up lines, not always verbatim but often close, that come from specific users.

      So it makes total sense that it would spit out vague, genericized Cobainims when fed the body of his lyrics. But it will never give you something with actual meaning except at random. Artistic merit is harder to determine - some lyrics are already meaning-opaque and impressionistic to begin with. For this song, I am supposing they ran the generator more than once, or spliced together multiple runs, to end up with something that didn't sound like total shit.

      The musical parts - the actual notes, chords, and so on - are basically already laid out as pseudocode in music theory. Pick a key at random, A through F, major or minor. There is a formula (the circle of fifths) that will tell you which chord progressions are "allowed", and scales to tell you which notes are "allowed" over those chords, and that covers the majority of songs you'll ever hear.

      None of this adds up to interesting music, but it could probably be developed to use in background music at the mall, for example. Once they find a programmer to wrap it all together for them nicely, for less than it costs to keep composers and musicians on hand. (Vocalists and thus a light-duty recording engineer would still be needed.)

      • The vocalists' days are numbered as well. You'll either be able to buy a deepfaked performance by someone who's dead and thus can't complain or demand royalties or get metoo'd or get assassinated, or just have a computer sing a part and then autotune and vocode it into whatever you want.

        • Wholly computer generated voices do have a place [youtube.com], but that locks you into a very specific niche of sound. It doesn't really capture the expression you find in Kraftwerk, Daft Punk, Lil Wayne, Young Thug, etc., and the variation between all of them. The vocalist underneath the vocoder has as much impact as the vocoder itself. Automating the drums and keys in popular music was easy, most non-musicians can't tell the sound of different drummers apart. But vocals have a much more identifiable human signature,

          • Deepfaking specific singers, especially ones with very distinctive voices like Dylan has yet to be perfected. But just getting a more generic 100% synthesized voice has been solved. Yamaha has been selling vocaloid as a commercial product since 2004. Hatsune Miku is a fairly well known marketing vehicle for the product, an entirely fictional pop star in a similar vein to an old Macross plot except this one hasn't gone full skynet yet.

            To my knowledge Vocaloid is mostly used by producers to create a filler vo

  • It sounds nothing like Nirvana (Score:3)

    by NotTheSame ( 6161704 ) on Sunday April 04, 2021 @05:20PM (#61236460)

    "It was a lot of trial and error," Over the Bridge board member Sean O'Connor told Rolling Stone. Once they had some musical and lyrical samples, the creative team picked the best bits to record. Most of the instrumentation you hear are MIDI tracks with different effects layered on top.

    The process of choosing which bits to stitch together (and in what order) had nothing to do with AI. There's far too much human involvement to make this an impressive feat. Also, the song is terrible.

    • Re:It sounds nothing like Nirvana (Score:4, Interesting)

      by hamburger lady ( 218108 ) on Sunday April 04, 2021 @06:33PM (#61236620)

      it doesn't sound like nirvana, no. it sounds like an okay nirvana cover band trying to create their own material that sounds like nirvana.

      that being said, give this bullshit a few more years and i'm sure it'll sounds pretty goddamn convincing.

    • Exactly.

      And I should have read that before I posted my comment saying the same thing.

    • Re: (Score:2)

      by AmiMoJo ( 196126 )

      Potentially useful though. Could be a handy creative tool for artists to use.

      My dream would be for AI to be able to bring back cancelled TV shows. Maybe get some new actors in for a bit of mocap and of course some writers, but have the AI fill in all the rest so that it's practical to do. Use it as an assistant.

    • Re: (Score:2)

      by hey! ( 33014 )

      There's far too much human involvement to make this an impressive feat.

      That's something we should bear in mind in all stories about "artificial intelligence", particularly ones about replacing people with AI. AI isn't magic, it's data processing, and is only as good at the data you put into it. That, in turn, is only as good as the people who select the data.

      The frontier between routine data processing and AI is a perennially receding one. There was a time when having a machine that could perform a sequence of more or less rote computations was amazing, and in that era it w

  • lacks ... a lot (Score:5, Interesting)

    by pz ( 113803 ) on Sunday April 04, 2021 @05:28PM (#61236476) Journal

    I read the article (I know, I know) and listened to the song (I know, I know, stop judging me). I remember when _Nevermind_ debued on the radio where I lived at the time (WBCN, for those old enough and in the right place at the right moment). It was played in full, the DJ saying, "this is going to be one of the most influential albums, ever," or something like that. He was prescient. It was amazing. Our aural and intellectual horizons were expanded in directions we didn't even know existed. It wasn't just a single, but a full album of that level of creativity. It lit our hair on fire.

    The single here, well, it sounds like pre-teens noodling in the garage trying to emulate their heros. It is neither a good song, nor a good imitation beyond blind copying. The only notes you hear are ones you've heard before from other Nirvana songs. There was no extrapolation to newness, only interpolation of what already existed. This is not a new Nirvana song, this is a low-grade copy of the geometric mean of Nivrana songs used as input. It's Nirvana without the soul of Cobain.

    Were I to attempt this sort of composition, I'd take the top N songs when Nirvana was recording as Type A input, the contemporaneous Nirvana songs as Type B input and build a network to (attempt to) encode the relationship between them. Then, feed the current top N songs as Type A input, and have it (attempt to) generate a Type B output. It was the contrast, the rejection of everything else, the passive-aggressive revolt, that made Nirvana so influential --- what would that contrast generate, now?

    • Re: (Score:2)

      by nadass ( 3963991 )

      what would that contrast generate, now?

      Ariana Grande doing covers in skits for late-night talk shows?!

      (I dunno, but that methodology seems quite clever.)

    • Re: (Score:2)

      by DThorne ( 21879 )
      While I get why they wanted to do a "stunt" like this, I don't think his entire catalog is enough data to generate a good result. I mean - it *might* have been, it was worth a shot, but I suspect that's the reason it's lacking. If you fed it the whole genre, it might have been better but of course is missing the point.

  • That really sounds like an early Nirvana demo, and the lyrics are *gasp* actually good.

    Thought provoking, to say the least.

    • Re: (Score:2)

      by vux984 ( 928602 )

      "and the lyrics are *gasp* actually good."

      Sure I mean the AI output:

      "I don't care/I feel as one, drowned in the sun,"
      "I won't share/ I make fun, listen to me run"
      "I can't care/ Drowned in the one, I smoke the sun" ...
      x1000

      And a human picked the best bits, and patched a song together.

      a pattern mimic made a bunch of patterns like the patterns that were input, where the input patterns were all nirvana songs.

      Then a human filtered through all the output, spent a lot more time tweaking the mimic settings, and inp

      • Color me simultaneously unimpressed and offended.

        Well, sucks to be you i guess.

        • Re: (Score:2)

          by vux984 ( 928602 )

          It sucks to be me because I'm neither impressed by yet another shitty so-called "ai" nor am I thrilled at how its being spun?

          Copy-cat and tribute bands have always existed, that's all this is. And nothing wrong with that, if they own it. And they don't go around dressing it up like "we're what X would have made if they were still around".

           

  • They have four videos (all posted two months ago). There is no info on who is being emulated, and comments are all turned off. https://losttapesofthe27club.c... [losttapesofthe27club.com] doesn't say. Dicks. So what are the other three?

    • It doesn't say, but they're relatively easy to figure out.

      Man, I Know = Amy Winehouse
      You’re Gonna Kill Me = Jimi Hendrix
      The Roads are Alive = Jim Morrison
      Drowned in the Sun = Kurt Cobain

  • Seachange (Score:3)

    by NicknamesAreStupid ( 1040118 ) on Sunday April 04, 2021 @06:22PM (#61236586)
    There are two ways to look at this. One is a critical review of the work as a piece of art. Two is a plausible product that can be marketed and sold. While not a nuanced example of the group's repertoire, it certainly provides nostalgia, sort of like Muzak giving an elevator ride a faint memory of the Beatles. This might become the 21st century Muzak, which was never a big business, nevertheless profitable. Vinod Khosla once predicted that music would eventually become a spontaneously generated service, specifically tailored to the individual's taste, mood, and context. Of course, this means AI will drive the supply chain from composition to performance. People will still be involved, somehow.

    • There are two ways to look at this. One is a critical review of the work as a piece of art. Two is a plausible product that can be marketed and sold. While not a nuanced example of the group's repertoire, it certainly provides nostalgia, sort of like Muzak giving an elevator ride a faint memory of the Beatles. This might become the 21st century Muzak, which was never a big business, nevertheless profitable. Vinod Khosla once predicted that music would eventually become a spontaneously generated service, specifically tailored to the individual's taste, mood, and context. Of course, this means AI will drive the supply chain from composition to performance. People will still be involved, somehow.

      Interesting point. I wonder, though, if copyright lawyers might not just respond by saying that the fact that the AI was trained on existing pieces by the original artist means that the product of the AI is a derivative work. If the AI does the work unassisted by humans, then I could see courts absolutely agreeing that the output is not just derivative, but a mechanical transformation, making it completely equivalent to the sum of the inputs, legally speaking. If some human judgment was injected in the pro

      • Your insight opens a can of worms regarding intellectual property and technology's effect. Today, there are still many human hands producing art using electronic technology, and there are many precedents regarding the evolution of creative work over the past 70 years. As technology accelerates, fewer hands are needed for more and varied content. Copyright is a good example, as this could touch on lyrics, score, performance, and now the source code of the tools used to produce it, e.g., AutoTune. Add thi

        • Someday, great musicians might also be great mathematicians, quantitative analysts, or scientists, an interesting mix.

          And the big record labels may pump out product that has no human hands involved at all.

          Probably not, because the listeners will still want to have a human artist they can gossip about. But that artist's role may become little more than a figurehead. Some pop stars have been accused of that in the past, but typically they at least had to have a little talent. That may end.

  • 27 Club (Score:3)

    by gosso920 ( 6330142 ) on Sunday April 04, 2021 @06:27PM (#61236604)
    Kurt "No, I don't have a gun" Cobain would have been 54 years old this year, if he hadn't died of lead poisoning.

  • "Other than the voice of Hogan, everything you hear in the song was generated by the two AI programs"

    That's a bit misleading (or a lot) ...

    "with the AI mostly outputting a wall of distortion instead of something akin to his signature melodies. "It was a lot of trial and error," ... the creative team picked the best bits to record.

  • IIRC the ending had the guitar playing singer who was improvising eventually having the competing AI in sync with him.

  • This is all fine and dandy but if the algorithm really wanted to create Nirvana's next single, it would automatically wire a large sum of money to IHeartRadio like everybody else does who wants heavy rotation.

  • Do I detect snippets from copywrited songs? Waiting for DCMA on this

  • Sounded good to me, and I am frankly amazed the AI did such a good job. Funny seeing the comments on how some try to convince others that their taste in music is not the proper taste. Don't understand what one gets out of that.

    • Snobber is a popiular compensation method for an inferiority complex.

      And if one crowd in all of history thinks they themselves are shit, it's the Nirvana target group, no offense. ;)

      Not excuse me, while I go look up the song eith the lyrics "*I'm a loser, baby so why don't you kill me, ...*" by that other band that I forgot. ;)

    • Everyone is too cool for school here at /.

  • trying to lead by imitating. ;)

    The point is that this could only ever be as much as the existing songs.
    While the whole point of a creative band is that they come up with *new* things.

    And it's not like if Cobain would be found to be still alive and would make a nee song, the reactions wouldn't be mixed either, precisely because it would be something new, not the fuzzy-lighted distorted nostalgic memory people have of Nirvana.

    Every time you rember something, you reprocess it with your currend memories, formin

  • Google A..blows it's brain out !

  • Math, math! (Score:4, Informative)

    by MS ( 18681 ) on Monday April 05, 2021 @07:27AM (#61237922)

    He was born in 1967 and would now be 54, not 52!
    He died in 1994, that's 27 years ago, not 30!

  • Comment removed based on user account deletion

  • For now, I will stick with AI/DC and Great Balls. At least that made me laugh. This just makes me kind of sad.

    Obligatory link
    https://www.youtube.com/watch?... [youtube.com]

  • Call me when an AI can make a song that sounds like Yes or Genesis from the 1970s.

  • Once they had some musical and lyrical samples, the creative team picked the best bits to record

    And therein lies the problem, I bet the full output generated was total garbage, it took a human to pick through that and selectively find something that sounded like Nirvana and form a coherent song. The million monkeys with typewriters and the works of Shakespeare comes to mind. They also mention having to fiddle with the input because the "AI" couldn't make heads or tails of it, so how much of the actual inp

Slashdot Top Deals

Get hold of portable property. -- Charles Dickens, "Great Expectations"

Close