ghost
commented
3 years ago I have come to the conclusion that Panako is not the tool for my use case. Reinhard Sonnleitner et al. solidify my conclusion. I do however find it strange that the github page says "The main algorithm, the Panako algorithm, has the feature that audio queries can be identified reliably and quickly even if they has been sped up, time stretched or pitch shifted with respect to the reference audio." As mentioned before the majority of results returned were completely wrong. I did tweak a lot of the settings and that did mostly nothing.
For anyone else dealing with compressed, sped up, stretched, pitched audio like me, I would suggest using the audfprint algorithm. I am getting unbelievably accurate results with it.
I'm having issues with matching. I'm using the default config. My use case includes recorded audio from FM radio stations. We know radio adds problems for audio fingerprinting and matching because of compression and distortion. However, the recordings are clear enough that I would expect at least a bit of a match. I've fingerprinted a few hundred studio WAV files into my database. Then I cut a song from a radio broadcast, the same song I have previously fingerprinted in my database with a studio WAV. I'm not getting matches and sometimes I see matches for different genres (E.g. A techno song matched to a hard rock song). To my ears the studio WAV I fingerprinted and the FM radio recording sound similar enough. Sure the radio has slight distortion and artifacts but there isn't a huge audible difference. I see a lot of settings in config.properties but I'm unsure what changes would help my situation. Any ideas? Thanks