Getting more than one fingerprint for each offset time?
meikidd opened this issue · 5 comments
Hi, this module is awesome! I am brand new to audio fingerprint, this module get me a good opportunity to learn how to code.
But I found that I got more than one fingerprint for each offset time, is it normal?
Sample Output:
time=34 fingerprint=238707
time=34 fingerprint=216179
time=34 fingerprint=347159
time=34 fingerprint=478295
time=43 fingerprint=601676
time=43 fingerprint=929303
time=43 fingerprint=1060439
time=45 fingerprint=132398
time=45 fingerprint=1049879
time=45 fingerprint=2360577
I am confused how to use these hashes to search in the database for matching hashes.
For example, if I have a record in database time=43 fingerprint=601676, does it mean the sample hit this db record?
or it requires a full matching time=43 fingerprint=601676, time=43 fingerprint=929303, time=43 fingerprint=1060439?
Hi, thanks!
Yes it is normal to (sometimes) have more than one fingerprint for each offset time.
You can see this on the Readme pictures: blue points are often linked to several gray lines. And you have one fingerprint per gray line.
The good strategy to consider if the sample is a hit is to count the proportion of matching fingerprints for a given offset in time between the reference track and the one you are analyzing.
I will publish code on this topic soon.
Good luck!
I will publish code on this topic soon.
Do you mean the implementation for section 2.3 in An Industrial-Strength Audio Search Algorithm?
Great to hear that, I am also looking for open source module for that. 👍👍👍
Yes the link to the paper is relevant.
You can also look at this python code to understand what you need to do https://github.com/worldveil/dejavu/blob/7f53f2ab6896b38cfd54cc396e2326a98b957d07/dejavu/__init__.py#L119
Awesome, thank you very much!!
For the record, here is how I implemented fingerprint search:
https://github.com/adblockradio/adblockradio/blob/master/predictor-db/hotlist.js#L127