Ask HN: How to create a small unique fingerprint of a 25h+ audio book?

5 points

7 years ago

I'm the author of m4b-tool (https://github.com/sandreas/m4b-tool) and I noticed, that there are not many sources for audio book chapters (especially non english ones) to tag my audio books correctly.

So i decided to create my own database for chapters, but at the moment i am struggling how to identify audio files correctly.

With ffmpeg extension chromaprint it is possible to create fingerprints of audiofiles:

  ffmpeg -i "input.mp3" -f chromaprint fingerprint.txt

This works nicely, but the longer the audio file, the bigger the fingerprint (which is reasonable). Since i would like to store the fingerprint in a database, the smaller it would be, the better.

But for a 25h+ audio book the process takes extremely long and produce a ±5MB (!) fingerprint file.

I can think of three ways to solve this problem:

  - Only take the first X Minutes of the audio file (fast, relatively small fingerprint storage, but inaccurate)

  - Hash the full fingerprint with e.g. sha512 (small fingerprint storage, but slow and is this accurate?)

  - Hash the X Minutes fingerprint (fastest, small fingerprint storage but most inaccurate)

Which would be the best way? Are there other ways, i did not think of?

Thank you

5 comments