Page 1 of 3 123 LastLast
Results 1 to 10 of 29

Thread: Bass localization: an experiment

  1. #1
    Senior Member
    Join Date
    Oct 2010
    Posts
    169

    Bass localization: an experiment

    How do we localize low frequencies? The primary mechanism comes from interaural differences in phase. From wikipedia:

    Quote Originally Posted by Wikipedia
    For frequencies below 800 Hz, the dimensions of the head (ear distance 21.5 cm, corresponding to an interaural time delay of 625 Ás), are smaller than the half wavelength of the sound waves. So the auditory system can determine phase delays between both ears without confusion. Interaural level differences are very low in this frequency range, especially below about 200 Hz, so a precise evaluation of the input direction is nearly impossible on the basis of level differences alone. As the frequency drops below 80 Hz it becomes difficult or impossible to use either time difference or level difference to determine a sound's lateral source, because the phase difference between the ears becomes too small for a directional evaluation.
    We often talk about other indirect cues influencing bass localization -- distortion and resonances are usually the most frequent cues brought up, because they're cases where higher frequencies have stronger and more well-defined localization mechanisms. However, our brains must associate the resonance with the bass note in order to bind them in a way that steers the localization of the bass note in the direction of the resonance. If we didn't bind these two sounds, we would perceive the resonance as something distinct from the bass note, and so our localization of the bass note should not be altered by the presence of the resonance. Therefore, when we talk about such cues, we're assuming that this binding process takes place across frequency, and is probably related to the correlation between the bass note and the resonance.

    I'd like to perform a simple experiment where I manipulate bass localization by varying the phase of low-frequency bandlimited noise in the left and right ears and measuring the subject's report of direction. This will establish a baseline localization, and I can get thresholds and all the normal junk. Then I want to introduce higher-frequency bandlimited noise (with a different phase relationship) on top of the low-frequency noise to try to disrupt or steer localization of the low-frequency noise. The higher-frequency noise will share the same temporal envelope as the lower-frequency noise, producing a correlation. I can vary this level of correlation and, if there's an effect, it should disappear with lower levels of correlation.

    Seem reasonable?

    I'm struggling through some of the details of the experimental design, and would love to hear any suggestions.

    Edit: one question I have is whether or not the temporal envelope defines the correlation, or if the correlation comes from the structure of the noise itself. I read a paper once on this subject (for higher frequencies and ILD), and I forget the answer, but I seem to remember that the temporal envelope was what was important.
    Last edited by MarkZ; 03-03-2012 at 02:51 PM.

  2. #2
    Senior Member
    Join Date
    Oct 2010
    Posts
    169
    So I began coding this today, and as proof of concept, I decided to take basic ITD thresholds on myself at various frequency ranges. Since this is really just pilot data, these are not amplitude matched, so it's impossible to discern whether thresholds changed as a function of differences in amplitude of my headphones. However, the results match fairly well with typical thresholds you see in the literature.

    Basic procedure: I played bandlimited white noise (1 octave bandwidth, ~60dB/oct slopes, 1dB passband ripple) in 1s bursts through my headphones, where the alignment of the waveform was randomly shifted between the left and right ears by the specified amount. In the graphs below, I'm characterizing it in terms of samples, based on a 192kHz sample rate. So a shift of 1 sample corresponds to ~5us (microseconds). My task was to judge whether the sound was coming from left-of-center or right-of-center with a mouse click (left or right click, correspondingly). I used a staircase procedure initially to determine roughly what the thresholds would look like (data not shown), and then performed the test with a list of shift values to bracket that estimate. I modified the list to include more extreme shift values for the extremely low frequency and high frequency tests, because it was much more difficult. It was a blind test, in that I was not aware of how much of a shift there would be in each trial, nor was I given correct/error feedback.

    Below I show individual psychometric curves at each frequency band tested. The filled squares are the raw data, and the line is a weighted Weibull function fit. I estimated thresholds based on where the Weibull intersected with the 75% correct level, and those are shown in the bar graph at the end.



    It's a little surprising that I was still able to localize based on ITD in the 1280Hz-2560Hz band, although my performance was definitely worse. So I might have just been using the lower part of that freq band. But what's interesting is that I basically couldn't localize below 80Hz unless I shifted the left and right ear by at least 50-75us. This confirms the idea that we're worse at very low frequency localization. But thresholds were still much lower in this band than I expected (even though thresholds at the intermediate frequencies are about what you see in the literature).

  3. #3
    Senior Member
    Join Date
    Oct 2010
    Posts
    169
    The next step will be to compute localization in the 80-160Hz range in the same way, but in the presence of higher frequency noise. The higher frequency noise will be neutral (no left/right bias), to see if there's a masking effect of any kind.

    After that test, I'll randomly add left/right bias to the higher frequency noise to see if that steers perception of the lower frequency direction. It probably will.

    The real meat of the experiment will come later, when I introduce a temporal envelope to both noises, and see if I can reduce the effect of the higher frequency noise on localization. Again, the idea is that bass localization can be steered only when it's correlated with the high frequency signal.
    Last edited by MarkZ; 03-03-2012 at 08:42 PM.

  4. #4
    Big Daddy Chad's Avatar
    Join Date
    Aug 2010
    Location
    East-Central Illinois
    Posts
    1,570
    I have a hard time with subwoofer placement at times, especially with extreme left or right bias.

    I generally use sizeable studio monitors, was Urei 809a now JBL 4412. I generally let them play full range then bring the sub in where they start to naturally roll off or just a TOUCH higher. This is pretty low to be honest.

    My first rig with the Urei monitors had the sub under the left monitor, in a corner. figured that these extreme low frequencies would be "omni-directional" and the corner would help. I just could not make it right at all. It was ALWAYS bass heavy on the left, or so it seemed, I don't know if I could attribute this to mutual coupling or my next point. I moved the sub to the center and was happy for many years that way.

    My newest endeavor has a set of 4412s flown, the sub is right in front of my pant-legs, off the floor 4" and isolated (for a temporary solution.) When I have the sub off the image in mentally high, as I expected, adding the sub pulls it down, this starts to reinforce my concept that it's merely tactile and possibly my sub left situation was biased by having the impact on the left side of my body more. This effect is such that I can now barely have the sub active, as in it's WAY down from the rest of the system but that TINY amount of tactile sensation moves the image. This is good BTW.

    So, being just one step ahead, and honestly I was going post this but since you started the ball rolling...... I did this a couple weeks ago.

    I wanted to remove the music from the subwoofer to see if it's was related to the source or if it really was tactile. I took the sub output and ran it into the sidechain of a gate, the input of the gate was pink noise, the output was filtered and sent to the sub, I was basically gating noise with the original signal to remove any original signal from the subwoofer. I fired it up, barely cracked the sub open and bingo, the "mental" image moves with gated noise.

    The reason behind doing this is that I'll soon, hopefully have a pair of IB subs mounted in th ceiling between the flown monitors. I like the tactile sensation that just having a tiny amount of LF in front of my "short and curlies" does. And I'm not joking when I say it does not take much at all. I was afraid of losing that so I pretty much decided that I needed something down low, small. so that Klipsch 10" subwoofer I scored as a giveaway will work just fine and I'll be running a system with multiple subwoofers.
    Last edited by Chad; 03-03-2012 at 08:56 PM.

  5. #5
    Senior Member
    Join Date
    Oct 2010
    Posts
    169
    It sounds like gating maintained the envelope of the sub input. So even though you changed the signal, it probably coarsely resembled the input, which might have been enough to couple it to the rest of the music. What happens to the image if you gated it with the sub input from another song? Aside from it sounding like shit, my guess is that our brains wouldn't be able to correlate the two, and it wouldn't have the same effect on the image.

    Edit: My aim isn't to show that bass is omnidirectional. It probably isn't, and even in my 40-80Hz test I could perceive left/right, even though it was much more difficult. What I think is going on is that when the directionality is weak, that's when other cues take over. And so very low frequencies, which are weakly directional, could potentially be dragged around by the strongly directional frequencies.

    There's a hint that this might be the case from experiments where visual input can alter the perception of the source of a sound. This is the ventriloquism effect, I know. But when you take that visual input and blur it, so that the visual cue is much weaker than the directional sound cue, you then have the OPPOSITE effect -- the sound steers the visual percept.
    Last edited by MarkZ; 03-03-2012 at 09:14 PM.

  6. #6
    Big Daddy Chad's Avatar
    Join Date
    Aug 2010
    Location
    East-Central Illinois
    Posts
    1,570
    I totally agree on all points. One thing about the Urei monitors is that they are flat out ugly and imposing looking, they LOOK like they will hurt. Funny how the "sound" changes when you place some acoustically transparent scrim over them ;)

    I'll play with noise some more, both another song and just bringing it in filtered, I'll do this by lightening up the gate A LOT so that it's not there when the track is silent.

    The mention of omni-directional was just as a history reference as to what I've been thinking/doing in my personal setting... since.. like... 1998 when I became a home-owner.
    Last edited by Chad; 03-03-2012 at 09:25 PM.

  7. #7

  8. #8
    Senior Member
    Join Date
    Oct 2010
    Posts
    169
    Nice find. This seems to at least confirm that I'm on the right track, re: correlation. I haven't read it closely yet, but it looks like they find that when they decorrelate the temporal envelope of the two sounds, their lateralization can be discriminated. But not when the temporal envelopes are correlated. That's exactly the hypothesis I've put forth in this thread. Although I still plan to extend the experiment to lower frequencies, because I also suspect that the weaker cue is disregarded in a winner-take-all fashion.

    After a quick glance, it also looks like they show that discrimination is aided by separation in frequency. I've seen similar things when I ran the experiment by introducing ITD=0 higher frequency noise on top of the low frequency signal. It effectively masked the localization of the low freq signal, but this effect was reduced when I separated them in frequency.

    Both of these findings together would have important implications when it comes to the "up front bass" topic. One of them might be that time alignment is more important than phase coherence. Another might be that shallower slopes, which should enhance overall correlations and reduce the separation in frequency, might improve the illusion of up-front bass. Something to think about.

  9. #9
    Big Daddy Chad's Avatar
    Join Date
    Aug 2010
    Location
    East-Central Illinois
    Posts
    1,570
    I use pretty shallow slopes, find it to help.

  10. #10
    Junior Member
    Join Date
    Feb 2012
    Posts
    1
    Not OT of localization, but I think this Thiel FAQ supports some of your "up front bass" thoughts.

    Even if it doesn't totally support it, I found the explanation helpful thought provoking

Bookmarks

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •