4 ms·

So fundamentally just a ton of locality sensitive hashing, cleverly chosen to match the domain. Still wild that they shipped this in the flipphone era.

• joshgachnang 3y ago
  Incredibly done in 2003. There's a python library called dejavu[0] that does similar hashing. It mostly works to make your own Shazam (assuming you can source the data to hash!) [0] https://github.com/worldveil/dejavu https://github.com/worldveil/dejavu
  • bequanna 3y ago
    Only issue I’ve had when playing around with Dejavu is the query speed slows linearly as the library size increases.
    • joshgachnang 3y ago
      Yup, agreed. Currently at 4 billion hashes and matching takes 30-120s. I haven't put effort into speeding it up yet because I'm using it for a trivia contest and we have a couple hours to figure out the clips.
  • pan69 3y ago
    There is also this .NET project: https://github.com/AddictedCS/soundfingerprinting https://github.com/AddictedCS/soundfingerprinting