My failure... I wanted to say high directivity speakers !
I already did some measurements comparing toe-in and straight position : the speakers crossed in front of listening place show better phantom image stability
I already use both ITD and ILD for the calculation. Contrary to the thesis, I have no problems with noise, I don't want to find real sound positions : I want to estimate phantom image. So I can work with generated signals (sweeps, aso).
Interesting that you also did something similar.
I analyse ITD but as it is a comparison between setups, I don't need to get exact values so no need to add a head diffraction model, I think.
I tried wavelet transform but it didn't get good pictures !
For the bandwidth, I will make trials with ERB : but the difficult part will be correlation between measurements and listening.
For the ITD, I use IACC, it is also easy to calculate and seems to give stable results.
Another question :
the estimated source is at the mean position given by ITD and ILD estimations. It works when ITD and ILD are opposed : this is the well known time-intensity trading. This also works when ITD and ILD give identical direction cues (the natural way).
So in the software, I supposed that the estimation would work in-between....
Last edited:
Not at all. To do this you would need a dummy head. And probably also use other parameters such as HRTF.
My purpose is easier : I record separately impulse records of left and right speakers. From those IR, I can calculate ITD (from IACC) and ILD for each frequency band. Combining both gives me frequency dependant localization of a theorical phantom central source. Idealy, this source should be central over all frequencies. This is not the case in real measurements due to asymetrical speaker responses, position and room.
Interestingly, you can see the effect of a non central listening position : the phantom may collapse more or less quickly to the nearest speakers, this depends on factors such as loudspeakers directivity. Ie, you may measure if low or high directivity is better for spatial reproduction
For a transparent head model both ear signals are the same.
yes another possibility
I don't measure influence of the head. I measure ITD and ILD due to loudspeakers and placement in room. An ideal setup, perfect speakers in an anechoic room, would give a perfectly central phantom image at all frequencies.
The independant measurements for L and R would allow to use a head model to estimate ITD and ILD as if there was a head, but I'm not sure that it would bring something.
This is what the software on my web site was designed to do, but the model of image location was very simple. It would be interesting to compare my results with yours. We both found that the center image was stabilized by crossing the speakers in front of the listener. Are you using real speaker polars? I'm not clear on the model input.
No, I use no model for the speakers, I use real measurements of IR.
I'd also like to compare both software and localisation model.
For my soft to compare stability of phantom image, I need to get :
- IR of left speaker and IR of right speaker at 0° for speakers facing listener
- same with some lateral move (30 or 50cm ?)
- same central position but with toe-in
- same toe-in and lateral move
but you can also download the soft (it's free here : http://www.ohl.to/audio/downloads/pops.zip) and try for yourself. The only tricky part is that Octave has to be installed, see setup page of Pops.
Sorry for delayed response - busy again. So you measure the impulse response at two point in the room separated by the distance between the ears? Then you use a calculation to find the IACC from which you get the ITD and the ILD? You use these to determine the image location? Is this all correct?
I would have to agree with a comment made earlier, that this approach is based on HRTF results and as such must have a head in place to get meaningful data. When I did my image software I used results posted by Blauert which were made as a composite that already had the head diffraction data included.
Some years ago I used a bowling ball to get decent data for head diffraction. It was actually reasonably close to the real data, but with a bowling ball the results could also be compared to numerical models. Used bowling balls are dirt cheap, almost free. Any object in the middle of the two IRs would be better than nothing.
I would have to agree with a comment made earlier, that this approach is based on HRTF results and as such must have a head in place to get meaningful data. When I did my image software I used results posted by Blauert which were made as a composite that already had the head diffraction data included.
Some years ago I used a bowling ball to get decent data for head diffraction. It was actually reasonably close to the real data, but with a bowling ball the results could also be compared to numerical models. Used bowling balls are dirt cheap, almost free. Any object in the middle of the two IRs would be better than nothing.
Actually, it is not very absorptive.
BTW, a bowling ball is not all that bad. We used to use one when calibrating set ups in a lab setting.
If I need a basic dummy head, I use a nicer solution : a hairdresser manikin with mics outside or in the ear entrance.
Then, if I move the mic another position, ie 50cm left, I can see the variation of ITD and ILD, and so phantom position variation due mainly to the directivity of the speakers.
with mics outside or in the ear entrance.
Not exactly : I measure separatly the impulse responses of L and R at listening position, same place for both mesurements. From those IR, I do a smoothed FFT to get ILD and I calculate IACC that gives me ITD. In this case, both ILD and ITD are only due to assymetrical L/R responses of loudspeaker+room. Those ITD and ILD give me the frequency dependant phantom position.
Then, if I move the mic another position, ie 50cm left, I can see the variation of ITD and ILD, and so phantom position variation due mainly to the directivity of the speakers.
Then you are doing the same thing that I am doing, but using a different m del for the image perception. Although
is not quite clear what you are doing. How is the phantom position "frequency dependent"? We won't perceive different phantom locations, just one - blurred maybe, but that's something else.
It depends on what you're doing : here it was for a recording with a middle sound barrier, a kind of Jecklin disk, with the ear canals blocked with absorbent cotton.
Back to the topic, I now understand that my terminology is confusing. I should maybe speak about PITD and PILD : phantom image time difference and phantom image level difference, and not use "interaural"
The standard setup for localisation with a real or a dummy head consists of one source and two receptors (both ears). Here I have inverted the setup : two sources (two loudspeakers) and one receptor.
It indicates that some cues may be contradictory : ie frequencies lower than 2kHz are 10 degrees at the left and higher frequencies are at the right. Even if you "finally" localise to the center, this blur is ideally to be avoided. And this blur is frequency dependant.
My soft should help to estimate if the phantom image is rock solid or if it is blurred.
I think your approach seems to make sense. If the drivers are not well matched, PITD and PILD will not match, and this will be shown as a skewing line, is this correct? If they are well matched, when you shift mic location, PITD and PILD should shift in similar manner. If not, most likely the polar response of the speaker may not be ideal.
I am curious though, how would one distinguish between a phase problem vs amplitude problem?
I am curious though, how would one distinguish between a phase problem vs amplitude problem?
This is correct and easy to show with hand-made impulse responses
Not an easy task because the well known trading ratios between ITD and ILD depends on signal ! I use default ratios of 0.02ms/d° and 0.5dB/d°. But I'm not sure that an amplitude problem can allways compensate a phase problem or the inverse.
I think an interesting experiment would to see how a tool like this corresponds with what people hear in terms of image focus. My assumption is that for a certain radiation and phase pattern, one could use this kind of method to determine an optimum setup in a room. It could also be used as part of a cyclic process to improve room acoustic characteristics. I think I will give it a try and see what I get. I probably might do it in different rooms using a same system as well.
Can I use one sound card as output and another sound card as input with this software?
Can I use one sound card as output and another sound card as input with this software?
Last edited:
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.