search

worldveil / dejavu

Audio fingerprinting and recognition in Python
MIT License
6.45k stars 1.44k forks source link

How to interpret results? #233

Open crtez opened 4 years ago

crtez commented 4 years ago

I was wondering if there was documentation on how to interpret each individual metric from a fingerprinting task--like for example, what is the difference between input_confidence and fingerprinted_confidence?

Thanks!

subins2000 commented 3 years ago

+1. I want to know how to interpret the results too. Like to find the song the most matched and how good of a match it is

georgiecasey commented 3 years ago

According to comments here: https://github.com/worldveil/dejavu/blob/acca344f23a1e2e98e01509e2c020f90f5fd22ba/dejavu/__init__.py#L211

input_confidence is Percentage regarding hashes matched vs hashes from the input.
fingerprinted_confidence is Percentage regarding hashes matched vs hashes fingerprinted in the db.

So input_confidence is what percentage of the input MP3 was matched and fingerprinted_confidence is what percentage of the entire song in the DB was matched. So input_confidence is really the value you care about and you want to get it as close to 1 as possible. fingerprinted_confidence is always going to be close to 0 unless your input MP3 is the entire song.