Re: Defeating Music Identification Apps - Adding Gaussian Noise?
Ivan K <ivan_521521-/[email protected]> Sun, 19 Apr 2026 23:38:06 +0000 (UTC)
| Newsgroups | gmane.linux.audio.users |
|---|---|
| Message-ID | <[email protected]> |
A belated thank you to all those you responded to my post! Paul Davis <[email protected]> writes: > > Did you see Benn Jordan's video on this (or an adjacent) > topic? > > https://www.youtube.com/watch?v=3DxMYm2d9bmEA I did, which is where I got the idea. On a related note, that video was taken down recently. "Jeanette C." <[email protected]> writes: > > it would be easy enough to do in Csound. At the very > least, I could send you a file to create as much Gaussian > noise as you want. Specifically mixing it with audio in > mp3 or the usual sndfile-supported formats is also easy > enough. Yes, I was thinking of this. I have not coded with Csound for 10 years or more, but I could get back into it. Fons Adriaensen <[email protected]> writes: > > For your quiz I assume you want to test if your students > can identify a particular period or style rather than the > actual title of the piece. > > The way to do this is to force them to motivate their > final choice. Instead of having just one multiple choice > between a number of periods or styles, also have them > identify features: > > - complexity, > - type of harmony, > - orchestration, > - temporal structure, > - .... > > providing a MC item for each of these. Actually, I do this, asking the students to identify style elements as well. However, I teach Community College and I have a wide range of students, some who could handle such a question, but many are quite frankly barely ready for college. For many students, I have to accept any vague answer that is close, or I would have to fail a good 70% of the class. Chris Caudle <[email protected]> writes: > > On Wednesday, April 8, 2026 10:28:40 PM Central Daylight Time Ivan K wrote: > > > > Also I have read how one can add a bit of gaussian noise > > to a sound file, and while to the human ear, the music > > sounds identical, adding this noise can interfere with > > or prevent these identification apps from identifying > > the music. > > An easy way to think about this is consider what would > happen if you played music on a device with slightly worse > signal-to-noise ratio, which is what additive Gaussian > noise would be, just a slight increase > in white noise. This is an online class. The students take the quizzes in a browser and play the examples on whatever speaks and/or headphones they have at home. --- One thing that I have learned is that Shazam (the most popular phone app that recognizes and names pieces of music) does so by consulting a library of commercially available audio, as Shazam does not just list the name/composer of a piece of music, but displays the CD image as well. And so, if the recording has not been commercially released, that is, I get the recording from an ensemble that posts their performances on YouTube, Shazam fails to recognize such recordings, even with incredibly famous pieces like symphonies by Beethoven or Mahler symphonies). I experimented with compressing the audio these files, but found that Shazam could still recognize them. Recently, I just used ffmpeg to mix some some gaussian noise with these audio files, but Shazam was able to still identify the piece of music. --- Thank you for your feedback.