Fingerprinting Audio With Differing Sample Rates

[rstokes92]

I have noticed that this doesn't really seem to work if the fingerprint in the database was taken from audio sampled at a different rate than the audio used to generate the test fingerprint. This is also mentioned in other issues (https://github.com/worldveil/dejavu/issues/94, https://github.com/worldveil/dejavu/issues/62) but never really discussed why or what the best practice is for addressing.

I guess the most practical approach is for everything in the database to be fingerprinted at the exact same sampling rate? What about if this is not possible?

Would be super interested to understand the reason for this given the spectrogram takes sample rate into account.

[alic-xc]

I think the most respected solution is to downsample it to fit the rate of the fingerprinted audio