3 ms·

I'm somewhat surprised that Google doesn't track public domain works and see them as such -- it seems like a perfect honeypot to distinguish copyright trolls (o

I'm somewhat surprised that Google doesn't track public domain works and see them as such -- it seems like a perfect honeypot to distinguish copyright trolls (or just over-aggressive studio/record-company algorithms) which they could regulate or push back against. I.e. public domain works are a good input to their own AI -- why are they leaving that on the table?

TonyTrapp 5y ago
A specific performance of a public-domain work may still be copyrighted. The problem is that their algorithms are not really tuned to distinguish different performances of the same song - on the contrary, I think it is safe to assume that their algorithms try to find as many similiar-sounding matches as possible, so that e.g. a song playing in the background on a radio is still recognized, even though it obviously sounds very different compared to a direct recording of a song.
- mod50ack 5y ago
  That's decidedly not the problem here. They didn't confuse my recording for a copyrighted performance. The claim was that I used a copyrighted melody in an original performance --- but the melody is public domain.
  - TonyTrapp 5y ago
    Out of curiosity, what were those four strikes then, if not for four different copyrighted performances of the work? From my understanding of how ContentID works, matches are always done against recordings provided by copyright holders, and not against abstract melodies.
    - mod50ack 5y ago
      It's an opera with various songs. For four of these songs, content ID reported a melody match with a composition. Keep in mind that compositions and performances have separate copyrights and often separate owners! Compositions have to be registered separately in content ID. https://support.google.com/youtube/answer/2822002?hl=en https://support.google.com/youtube/answer/2822002?hl=en What happened was that each song was correctly identified by its melody and matched with an entry in the composition database, even though the compositions are clearly public domain.

TonyTrapp 5y ago

A specific performance of a public-domain work may still be copyrighted. The problem is that their algorithms are not really tuned to distinguish different performances of the same song - on the contrary, I think it is safe to assume that their algorithms try to find as many similiar-sounding matches as possible, so that e.g. a song playing in the background on a radio is still recognized, even though it obviously sounds very different compared to a direct recording of a song.

mod50ack 5y ago
That's decidedly not the problem here. They didn't confuse my recording for a copyrighted performance. The claim was that I used a copyrighted melody in an original performance --- but the melody is public domain.
- TonyTrapp 5y ago
  Out of curiosity, what were those four strikes then, if not for four different copyrighted performances of the work? From my understanding of how ContentID works, matches are always done against recordings provided by copyright holders, and not against abstract melodies.
  - mod50ack 5y ago
    It's an opera with various songs. For four of these songs, content ID reported a melody match with a composition. Keep in mind that compositions and performances have separate copyrights and often separate owners! Compositions have to be registered separately in content ID. https://support.google.com/youtube/answer/2822002?hl=en https://support.google.com/youtube/answer/2822002?hl=en What happened was that each song was correctly identified by its melody and matched with an entry in the composition database, even though the compositions are clearly public domain.