I want to train a model for a musician based on piano covers they have recorded of popular songs and therefore own, but because of copyright infringement they still pay royalties to the writers of these songs. Curious if this is legal or not – I would presume not since the recordings are not being monetized in this scenario but rather fed into a model, and the recordings are still owned 100% by the musician licensing them to me.

  • IgnisIncendio@alien.topB
    link
    fedilink
    English
    arrow-up
    1
    ·
    10 months ago

    What jurisdiction? IANAL, but in countries with TDM exceptions like Singapore or the EU, usually the training material itself must be legal (e.g. no hacking).

  • Blackliquid@alien.topB
    link
    fedilink
    English
    arrow-up
    1
    ·
    10 months ago

    Depends on where you are: in Germany there is no fair use for example. Everything that does not have a permissive licence is illegal.

  • f10101@alien.topB
    link
    fedilink
    English
    arrow-up
    1
    ·
    10 months ago

    Yes, the recordings are 100% owned by the musician.

    The songs themselves though, are not. If your focus is to generate new songs, then while there’s a good chance you’re in the clear, it is still an unanswered question. You’re essentially taking a bet that it’s legal.

    If your focus is to be able to do a style transfer, i.e. turning music into something that sounds like it was performed by your pianist, then use of the data becomes something that is more easily covered under existing fair use.

    I would suggest it may be something of a moot point though: can you actually build a useful musical model with this little data?

    • SuperwhizAJ@alien.topOPB
      link
      fedilink
      English
      arrow-up
      1
      ·
      10 months ago

      great points. the focus is so he can create more covers with the model. so the latter.

      would the fair use norm be that its transforming the original songs/being trained on transformations so it would be all good? the original songs would be used during training as input to the model, but any generations would still have royalties be returned to the songwriters should the musician intend to monetize them.

      re your last point: this musician has a lot of data. 30+ hours :P

  • JustDoinNerdStuff@alien.topB
    link
    fedilink
    English
    arrow-up
    1
    ·
    10 months ago

    We need to back all of this up- What is the purpose of you training this model? Are you going to make money off of it? Because from a practical standpoint, if you are not going to make money off of it, you are completely safe to do whatever you want. This coming from a 20 year professional artist in film and tv where we deal with copyright daily when sourcing and using found assets, both at work and at home. The only way you actually get in trouble for stealing art, is when you make money off of said stolen art. For this to happen, the person you stole music from would have to find out you did in the first place, which is easy to avoid. Then if they were to sue you, they’d have to PROVE DAMAGES to take your money. They’d have to show a judge that using your model, containing their work, you have made money, and that money opportunity was lost to them. Would someone be able to do those things with how you are planning on training and using this model? I don’t know your plans, but I’m guessing absolutely not. So in short, assuming this is a hobbyist project, you’re fine.

    • SuperwhizAJ@alien.topOPB
      link
      fedilink
      English
      arrow-up
      1
      ·
      10 months ago

      Okay yes, technically it would very difficult to track and no one would probably even know the musician is using such a model. But as an academic, I’d prefer to do legal things only :P especially if I want to use results in a publication down the line.

      And yes, I’d make money off it. would charge $x/full piano cover request.

      • JustDoinNerdStuff@alien.topB
        link
        fedilink
        English
        arrow-up
        1
        ·
        10 months ago

        This is a bit of gray zone then. To be honest, I’d probably just do it because every major ai company in the world is already doing way worse than what you’re doing, and making billions. The reality is their actions are creating legal precedent in your favor. Not LITERAL legal precedent, but abstract in the sense that every day they do business and DON’T get sued, we trend towards their actions being considered legal when it comes to blows later. It’s an open secret that landmark laws and cases are often dictated simply by what aligns with the interests of the current government administration, not necessarily what is most fair to everyone, or even constitutional. If the US government is doing millions/billions of dollars worth of business with ai companies, and these companies are doing ethically ‘gray’ training, it’s a pretty strong bet the judgements will be rigged in favor of Ai. I’m not telling you to betray your morals, but I’m just saying the wind is blowing that way, and it’s hard to sail against it.

  • Ronny_Jotten@alien.topB
    link
    fedilink
    English
    arrow-up
    1
    ·
    10 months ago

    There are several things involved, copyright of the song by the composer, copyright of the particular performance by the performer, and mechanical reproduction rights of the recording. The musician may be able to grant you rights to the performance and recording, but not to the composition.

    It may also depend on what you’re doing with the model. If you analyse the composition, and create new songs in the style of the composer, that’s more likely to infringe their copyright, than if you’re analysing the performance style of the musician, and creating renditions of other songs in the musician’s style. If you’re doing scientific research, it’s likely to enjoy a fair use exception. If you’re monetizing it, and it takes away income from the composer, then not. Even if you’re not monetizing it, there’s still a possibility to get a “cease and desist” order from the copyright holder if you don’t have a fair use exemption. It’s complex, and depends on the laws, court decisions, and international treaties. You won’t likely get a reliable explanation from Reddit armchair copyright experts, including me.

    The legality of AI and music is still up in the air to some extent. We know from the Heart on my Sleeve track that it’s not just a fair-use free-for-all as some Redditors here suggest.

  • theLanguageSprite@alien.topB
    link
    fedilink
    English
    arrow-up
    1
    ·
    10 months ago

    This is not legal advice. Please consult a copyright lawyer about this and take their advice over anything on the internet.

    However, there is a such thing as transformative fair use, where a remix that incorporates part of a work while altering it in order to create new meaning or purpose is not copyright infringement. Practically speaking, in court this often comes down to how different the allegedly infringing song is from the original. So ultimately there isn’t a single answer as to whether this is legal or illegal, with each song produced through inference it comes down to how likely it is for one of the song owners to sue and how likely you are to win.

    All the more reason to consult a copyright lawyer about it to get a better idea of your chances/options