Does anyone know of a local audio upscaler? Preferably Android based.

        • harald_im_netz@feddit.de
          link
          fedilink
          arrow-up
          11
          ·
          9 months ago

          Hello, Audio Engineer with some little knowledge regarding AI here.

          What you think of is restoring frequencies, this is possible, and commonly used in plugins for audio restaurization. I might be mistaken, but this does not improve the bitrate, but the perceived quality (which is still lossy).

          I don’t think that there is a real interest to upscale quality (not perceived quality), especially for longer (> 1 minute) material.

          • sabreW4K3@lemmy.tfOP
            link
            fedilink
            English
            arrow-up
            1
            ·
            9 months ago

            It’s funny you say that. I, as most people bought a bunch of CDs back in the day and ripped a bunch before I gave up my CD drive. At the time, storage was expensive and so I did what I could at the time with MP3. As storage gets cheaper (though not cheap enough for me to go lossless), I’d like to be able to upscale my music while keeping a similar file size and have my collection mature with me until storage becomes cheap enough for me to go lossless.

            I can’t be the only person who’s thought of this.

        • MangoPenguin@lemmy.blahaj.zone
          link
          fedilink
          English
          arrow-up
          6
          ·
          9 months ago

          It wouldn’t be the original audio, the AI would just be making up new content to fill in the blanks like it does with a photo.

            • Zoom@lemmy.studio
              link
              fedilink
              arrow-up
              2
              ·
              9 months ago

              If you’re OK listening to a derivative work of your input. Otherwise, it’s bad.

              • sabreW4K3@lemmy.tfOP
                link
                fedilink
                English
                arrow-up
                2
                ·
                9 months ago

                Hmmm. I feel like this is one of those long-term studies that would be quite exciting? Am I wrong to be a little bit excited about programs learning how to guess correctly what should be where and subsequently how things should sound?

      • sabreW4K3@lemmy.tfOP
        link
        fedilink
        English
        arrow-up
        2
        ·
        9 months ago

        Can AI or machine learning not do in the same way that it does with pictures?

        • Atemu@lemmy.ml
          link
          fedilink
          arrow-up
          9
          ·
          9 months ago

          It cannot bring back lost data. It can hallucinate something that is statistically likely given the context but I’m not aware of any tool which can do that to a useful degree.

          What’s the context? Why can’t you just get a better encode where the data isn’t lost?

            • Zpiritual@lemm.ee
              link
              fedilink
              arrow-up
              6
              ·
              9 months ago

              Old mixtapes and such can be noisy with hizzes, pops and such. It is possible to filter out those artefacts but thats removing stuff, just as digitally compressing audio is removing stuff. You can’t create data from nowhere for digitally compressed files and you can’t simply add back the hizz, noise, and pops to the mixtapes if you remove that.

    • aleph@lemm.ee
      link
      fedilink
      English
      arrow-up
      5
      ·
      edit-2
      9 months ago

      Going from 192kbps to 320kbps would be audibly negligible unless you used a really bad codec to begin with, in which case adding AI into the mix would likely just compound the problem.

      Probably not even worth it, tbh.