• Noxy@pawb.social
    link
    fedilink
    English
    arrow-up
    22
    ·
    1 day ago

    I was pretty enthralled by a lot of the stuff in this video. Kinda wanna try birdnet-pi now!

    a very minor nitpick: it’s not a PNG at all, it’s a lot fuzzier than being an actual image format. but I get that he’s gotta dumb down the video title so it’s not really a big deal

    • zqps@sh.itjust.works
      link
      fedilink
      English
      arrow-up
      22
      arrow-down
      1
      ·
      1 day ago

      The original file was a PNG. It stopped being a PNG when it was encoded as spectrogram of an audio file. Obviously in the bird’s memory the data is neither png nor any other machine-readable file format, but electrochemical signals.

      • ChaoticNeutralCzech@feddit.org
        link
        fedilink
        English
        arrow-up
        4
        ·
        23 hours ago

        And when recovered, the image resembles a drawing of a bird with quite a bit less detail, but also a somewhat horizontal line through the middle of it.

        I’m guessing one could make about 250 drawings that can be reliably distinguished in the bird’s call, storing one byte with decent reliability, which can be boosted at scale with Reed-Solomon. The “hundreds of kilobytes of uncompressed data” claim is ridiculous, I can “make” 4 MB of data by taking a 16MP photo of a 10-byte phone number.

    • SpruceBringsteen@lemmy.world
      link
      fedilink
      English
      arrow-up
      2
      ·
      1 day ago

      His other video on audio surveillance is eye opening. Stuff like that is relatively accessible to a layperson nowadays, it’s scary to think what’s possible on the cutting edge of things.