Forums

basic question about HRTF based sound localization

Started by Sylvia September 19, 2007
I have basic question about the HRTF based sound localization.Do the HRTF
based sound localization algorithms suffer from front-back ambiguity
problem?I am trying to use one of the HRTF algorithms to detect sound
source direction in azimuth in the presence of noise,sometimes,it detects
the  azimuth symmetrically in the back plane when source is infact in 
front planeand viceversa.What can be the reason of this front-back
reversal?
On Sep 19, 9:25 am, "Sylvia" <sylvia.za...@gmail.com> wrote:
> I have basic question about the HRTF based sound localization.Do the HRTF > based sound localization algorithms suffer from front-back ambiguity > problem?I am trying to use one of the HRTF algorithms to detect sound > source direction in azimuth in the presence of noise,sometimes,it detects > the azimuth symmetrically in the back plane when source is infact in > front planeand viceversa.What can be the reason of this front-back > reversal?
it's not because of HRTF that there is this front-back ambiguity, but in spite of it. if we had no ear pinnae and our ears where just little holes on the two sides of our head (or, perhaps, eardrum membranes flush with our head surface), then there would be complete front-back ambiguity. there would be *only* the interaural time difference (ITD, which happens to be the primary que we use to determine sound direction, giving us good left-right discrimination) which is what the Blumlein stereo model is entirely based on. drawing a straight line or axis through our head and both ear orifices, if some sound impinges upon the listener (without pinnae) from some angle relative to that axis, you can imagine a cone having the same coincident axis, that the sound source lies on. but that sound source could lie *anywhere* on that cone; front, back, above, below, and the interaural time difference would be the same. the *only* physical mechanism that exists that helps the listener know the difference between front, back, above, or below, are the ear pinnae (that are not symmetrically shaped regarding these directions). so if there is front-back ambiguity, it is *despite* the HRTFs based on the ear pinnae, not because of them. IMO. the pinna ques are more subtle (perhaps less "effective") than the ITD. the ITD is very solid physically and physiologically which is why, if you assume everything is in front of you and on the same plane, you have very good left- right localization. when sound sources move up or down or behind you, then your localization might not be as good. you might have difficulty discriminating it from something in front of you that has the same ITD. r b-j
r-b-j thanks
i understand the theory u explained.infact,it is the HRTF that helps us to
localize the sound infront or back.
but because of noise,it is possible the detection algorithm based on HRTF
makes wrong localization.But the question is,why i am getting most of the
wrong detections as the front-back symmetric version of the angle.for
example,if a source is infront at 0 degree,using hRTF algorithm,i am
getting it at 180 degree just behind the head.Does noise effect the HRTF
cues and modifies them in such a way that sometimes,we get front-back
version of the actual angle.

Thanks
Sylvia

On Sep 19, 9:25 am, "Sylvia" <sylvia.za...@gmail.com> wrote:
>> I have basic question about the HRTF based sound localization.Do the
HRTF
>> based sound localization algorithms suffer from front-back ambiguity >> problem?I am trying to use one of the HRTF algorithms to detect sound >> source direction in azimuth in the presence of noise,sometimes,it
detects
>> the azimuth symmetrically in the back plane when source is infact in >> front planeand viceversa.What can be the reason of this front-back >> reversal? > > >it's not because of HRTF that there is this front-back ambiguity, but >in spite of it. if we had no ear pinnae and our ears where just >little holes on the two sides of our head (or, perhaps, eardrum >membranes flush with our head surface), then there would be complete >front-back ambiguity. there would be *only* the interaural time >difference (ITD, which happens to be the primary que we use to >determine sound direction, giving us good left-right discrimination) >which is what the Blumlein stereo model is entirely based on. drawing >a straight line or axis through our head and both ear orifices, if >some sound impinges upon the listener (without pinnae) from some angle >relative to that axis, you can imagine a cone having the same >coincident axis, that the sound source lies on. but that sound source >could lie *anywhere* on that cone; front, back, above, below, and the >interaural time difference would be the same. the *only* physical >mechanism that exists that helps the listener know the difference >between front, back, above, or below, are the ear pinnae (that are not >symmetrically shaped regarding these directions). > >so if there is front-back ambiguity, it is *despite* the HRTFs based >on the ear pinnae, not because of them. IMO. the pinna ques are more >subtle (perhaps less "effective") than the ITD. the ITD is very solid >physically and physiologically which is why, if you assume everything >is in front of you and on the same plane, you have very good left- >right localization. when sound sources move up or down or behind you, >then your localization might not be as good. you might have >difficulty discriminating it from something in front of you that has >the same ITD. > >r b-j > >
"Sylvia" <sylvia.zakir@gmail.com> wrote in 
news:gvCdnfaOqKUbBmzbnZ2dnUVZ_gCdnZ2d@giganews.com:

> but because of noise,it is possible the detection algorithm based on HRTF > makes wrong localization.
Even with a perfect HRTF, localization is not perfect. There is a cone of confusion, so differentiating front/back can be tough. With a slightly off HRTF, sounds are localizable, but not necessarily externalized. Elevation is also a bit tougher than azimuth. -- Scott Reverse name to reply