This is a hypothetical question but please hear me out. I know that Audio Fingerprinting systems such as Shazam use perceptual hashing instead of cryptographic becuase a single bit flip due to how the audio was encoded or noise when the recording took place wouldn't match the clean hashed fingerprints of the audio at the database side but would it be possible to use a perceptual hash to find the features of the audio you wanted to record and then run those frequency peaks (sub-fingerprints) through a cryptographic hash? You would do the same at the database end on the clean version of the song and then surely some of hashes would match if compared? Or am i missing something obvious here. I know this would make it computationally more expensive & slower but was just wondering if this would be possible..
Asked
Active
Viewed 196 times
0
-
Have i asked this on the wrong forum perhaps? – Rob Aug 17 '20 at 19:58
-
Nobody? :( I know it might come accross as bit ambiguous but surely it's possible.. – Rob Aug 18 '20 at 08:23
-
Is the answer that using perceptual hashes don't have to be an exact match? As i thought even though you are using fourier transform on your audio sample to find many sub fingerprints (Peak Pairs), they have to exactly match the sub fingerprints of the clean database audio sample? – Rob Aug 18 '20 at 12:53
-
I think i know what the answer is lol. My way would work but the point is you have to use a perceptual hash in the first instance so makes my question redundant and you should just compare the perceptual hashes as that's what's been used to index the database anyhow... – Rob Aug 18 '20 at 13:24
-
Was i right? Feel so alone on this thread lol – Rob Aug 18 '20 at 18:17
-
No one? lol please tell me i'm not going mad ... – Rob Aug 20 '20 at 14:37
1 Answers
0
You definitely could. As long as whatever features you're extracting are fuzzy and noise-resistant, any hash function will do.
Obviously you'd prefer a faster hash with less collisions of course!
lollercoaster
- 15,969
- 35
- 115
- 173