When I am listening to music using head-phones, I find listening to stereo recordings rather tiring, so I am using a cross-feed-plugin (http://bs2b.sourceforge.net) which to my perception improves the spatial perception of the recording a lot (feels to me like having a panoramic perception instead of one that feels like having a pool cue stuck through both of my ears).

Obviously, I don't want to smash binaural recordings by applying cross-feed on top of them. So, this question came to my mind:

Is there any algorithmic way to guess whether a specific recording is binaural or not? Criteria that come to my mind:

  • rather easy: lack of cross-correlation -> stereo
  • difficult: consistent cross-correlation for certain "components" of the sound (wavelets?) -> binaural
  • vague: frequency envelope that is typical for head-transfer-functions -> binaural

Obviously, any such algorithm will have flaws because in fact any binaural recording can be interpreted as a strange stereo signal. And any stereo signal can be interpreted as a binaural recording in a very strange room, where the head is in the middle of a wall and each ear only hears part of the room.

Anyway, is there anything established in that direction?

  • 2
    I am unaware of any such work but you might have fun doing some original research. I suggest starting by comparing known stereo to binaural recordings, looking at phase shift vs. frequency and also direct delta between the two channels. You may notice some qualities that are present in true binaural recordings that perhaps don't show up in synthetic stereo. It will be difficult though. An easier option, honestly, may be to, say, put the word "binaural" in file names of recordings you know to be binaural, and do not apply the cross-feed to files with that word in the name.
    – Jason C
    Jun 17, 2014 at 20:34
  • 1
    Perhaps you could also look into existing research into how the ear processes sound, and develop a technique to determine how "spacey" a sound would be perceived (there is a lot of work on this subject, and probably some related programming work as well; start at en.wikipedia.org/wiki/Sound_localization).
    – Jason C
    Jun 17, 2014 at 20:38
  • @JasonC Thanks for the hints. I think, it's exactly the spaciness that one should try to quantify. That's what I mean with my comment to Rick T's answer. Jun 25, 2014 at 13:37

1 Answer 1


It is possible, however you would need to have the original signal. I use octave/matlab to create and test vocal signals. You need to do this for both signals separately
1) fft (Fast Fourier transform) on the signal

2) get all the frequencies, amplitudes and phases and place them into arrays

3) compare the data stored into said arrays to each other.

Note: the signal should be the same time duration and sample rate. This can be worked around but would require more programming. Also these files can be very large so a 64bit system would help, I use Linux

  • Ok, interesting thought, however the requirement to have the original signal is very hard. So, wouldn't it be possible to analyze the signal according to whether the amplitude relations and the phase relations of individual features are consistent with the binaural recording model? Jun 25, 2014 at 13:33
  • @Tilman Vogel You would have to have all three Frequency,phase, and amplitude. This can be done by doing an fft on the signal. I guess I miss understood, I thought that you already had the original signal. But you can also try phase modulation of the signal I've had success at getting a spatial feeling to the sound. Take note you want to change phase at the bit depth (more accuracy) of the signal to get what you looking for.
    – Rick T
    Jun 27, 2014 at 2:24

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.