7

During a presentation (like a quiz), I want to use some kind of music , but don't want listeners to detect them by sound-recognition programs (like Shazam or etc).

Is there any plugin, media-player, equalizer effect or anything, so listeners(humans) cant detected it (but as the same time, the music should practically remain same to human ears)..

T.Todua
  • 189
  • 1
  • 1
  • 7
  • 3
    pub quiz? I'd google for "get around youtube content id system". I doubt that shazam is better than the audio recognition software used by record companies to get their pound of flesh, if you can fool Universal, you can fool shazam – Some_Guy Oct 24 '17 at 17:58
  • 1
    I think a small timestretch on audacity will probably do it. Not posting as an answer because I don't know for sure – Some_Guy Oct 24 '17 at 17:59
  • If you're asking because you would like to circumvent a copyright enforcement system, then your question is not allowed on stack exchange. If not, then I'm a bit curious why you're asking. And why you want it to be real-time. – Todd Wilcox Oct 25 '17 at 02:19
  • 2
    @ToddWilcox , no i am not trying to deceive any system. I am trying to setup a quiz-like (not detectable) music to person,as a person shouldnt be able to "trick" me and find music using program. – T.Todua Oct 25 '17 at 09:34
  • Not what you are asking, but would playing it on guitar/piano make it unrecognizable for Shazam? – Karlo May 20 '18 at 17:49
  • @Some_Guy that clearly won't work because I shazam all the time at clubs where the DJ's music is time stretched and locked to a key, or the pitch is shifted +/- 10% and it works perfectly – Tim Davis May 20 '20 at 06:32
  • 2
    @TimDavis The comment you are replying to is 3 years old, audio recognition has come a long way since then, and that's true of content id on youtube too. In my experience to evade the algorithms these days people tend to use a combination of time dilation, pitch shift and strange phasing & stereo effects, and this really impacts the sound of the music much more (and is still not guaranteed to work) – Some_Guy Jun 29 '20 at 00:30
  • I’m voting to close this question because this is not about music practice or theory. – Dom Sep 18 '21 at 16:05

4 Answers4

13

In short, as @Some_Guy suggested, making a small time dilation might work.

Shazam made music recognition possible by generating "sound fingerprints" of small portions of an audio segment, and compare these fingerprints with those recorded in its database. To get rid of the recognition system, you have to make sure that any audio segment with the same fingerprints does not exist on Shaman's database. So it's either 1) Shazam doesn't have the music piece or any music with similar portions to yours; or 2) Shazam has the music piece, but somehow you managed to fuzz it so Shazam is obfuscated.

To generate the "sound fingerprints" of an audio clip, Shazam starts by splitting it into many short segments. By saying "short", we mean about a hundredth or thousandth of a second. Then for each short segment, Shazam picks out the most significant frequency ranges in it. The frequency ranges are then encoded (or in a programming term, hashed) and stored, along with a timestamp of this short segment. The hash and the timestamp together we call it a sample. Thouthands of samples are generated and then sent to Shazam's server, to find a matching piece of music in its vast database with billions of samples generated from millions of music pieces.

The matching process is quite straight-forward. Since the most significant frequency ranges of each sample are hashed, what Shazam tries to do is to find samples with the same hash values, and see whether these found samples can be timeline-aligned with those samples to match. A certain hash value can appear in many music pieces, but it's less possible for a sequence of hash values to appear in two different music pieces. The longer the sequence is, the less possible a hash-conflict might happen.

Some more facts about Shazam's music recognition algorithm:

  • The timeline aligning need not to be perfect. It doesn't matter if some of the samples does not match. Shazam scores all possible music pieces, and pick the one with the highest score as the result. As stated in their paper, from a heavily corrupted audio clip, they can find a match with only about 1-2% effective samples. So it won't work if you corrupting part of the music (which is also not acceptable to OP's requirement).
  • Shifting the pitch of the music might work, but since the algorithm is designed to calculate against frequency ranges instead of exact frequency points, you have to change the pitch by a large magnitude, which is definitely recognizable by the audiences. There are also some other filters which might be able to cheat Shazam, but it's still impossible to do that with an imperceptible change.
  • Shazam's algorithm can extract the transparency of multiple tracks mixed in an audio clip, so if you are mixing a Shazam-recognizable music with other sound tracks, it's still detectable.

It seems Shazam has a very strong algorithm which is impeccable, however there is one advantage we can use, the accuracy of time. A matching in Shazam is determined by an exact timeline alignment. With this characteristic, Shazam can even distinguish between two versions of a same song, or tell whether a singer is faking his live with lip synchronization. The system is accurate down to milliseconds, so a small amount of time dilation can wreck it down.

That being said, it's still technically possible to handle these kind of obfuscation, especially if it's a linear one, and if you are not facing Shazam, which is designed to recognize exactly matching musics. Not to mention Shazam's algorithm is initially published in 2003. 14 years later, there are much stronger algorithms to detect music similarities, such as machine learning.

You can read Shazam's paper to learn more about its algorithm.

hillin
  • 768
  • 5
  • 16
  • Excellent answer, wow! but I asked in general, I knew Shazam, however, which ones are more stronger systems? I also meant for those systems too,or anythng, that is music-recognition.. – T.Todua Oct 25 '17 at 09:39
  • 1
    Most of these services are derived from Shazam's algorithm and serve the same goal(to find out the exact music you hear in an anonymous context), so I think this answer applies to them too. However there are other systems made for different purposes. For instance, Youtube's content ID system is designed to identify copyrighted materials, although I'm not sure, I guess it's very possible that the system has an anti-obfuscate feature to handle time dilation. – hillin Oct 26 '17 at 02:37
  • Since you're doing it for a clip, you might be able to add some noise that would make it harder for it to be machine recognized, but not affect how it can be human recognized. Kind of like when Shazam can't figure out a song because it's got static, or there is too much background noise... but you can still hear it clearly enough to identify it. – Greg May 18 '18 at 15:35
  • @Greg background noise wont help, as said by HILLIN. – T.Todua May 20 '18 at 16:25
1

Backmasking the song will break Shazam for sure

Edward Anthony Gomez Gatchalia
  • 19
  • 1
  • To complete this answer, try explaining what backmasking is and why it will work in an edit to your post. As it stands, it looks a lot more like a comment than an answer. – user45266 Sep 20 '21 at 01:22
1

Keep same tempo and Try to Change key to +1.5 or +2. That should fool Shazam.

Richard
  • 82,618
  • 16
  • 186
  • 360
1

Many popular music radio stations (Contemporary Hit Radio) pitch music up as a standard, sometimes anywhere from +1 to +3 percent, so I would think the recognition would be designed to handle quite a bit of time or pitch error.

Glorfindel
  • 3,612
  • 5
  • 21
  • 36
Scott
  • 11
  • 1