dpwe/audfprint

does the --illustrate option work?

Opened this issue · 2 comments

db48x commented

I've been playing around with audfprint and am trying the --illustrate option to see how things work, and try to figure out why some matches are poorer than otherss

Here's what I see for one particular match that I'm interested in:
Screenshot from 2019-07-19 16-07-26

Do the fingerprints really cover so little of the audio, or it just being drawn with the wrong scale?

Where does the time axis fall relative to the match? This match is 1768.4 seconds into the reference file, and is only 0.2 seconds long (and yet it's not a false-positive).

db48x commented

After staring at the code for a bit, I think the hashes are scaled wrong. Testing some likely-looking scale factors certainly makes a better-looking plot:

Screenshot from 2019-07-19 17-19-24

Also, my query is 9.7 seconds long, so the time axis could be relative to the query.

db48x commented

After the fix:

Screenshot from 2019-09-22 18-25-35