After re-reading the article, Fixing the Phantom Center, I realize that my last entry was a guess at what was meant. It seems that the bottom line is that you add decorrelated reverbs to each channel that are so short term that they are not perceived as reverbs, but cause comb filter cancellations of the phantom center to be largely filled in by the various non-correlated delays. Much like how room reflections do that. The paper recommends BW limiting the processing to above 1kHZ so timing cues in the lower midrange aren't scrambled, which they are anyway due to inter-aural crosstalk. But for headphone compatibility that makes good sense.
But if the delay times are that short as to not be perceived (2mS was mentioned), won't that cause the distance between the nulls to be perceptable by a person sitting near the L or R speaker? A sort of frozen phlanging effect? The article mentions that the improvements are only significantly perceptable in relatively dry environments. Maybe if this same technique was used with an actual center speaker, it might be a significant improvement. 3 different effectively decorrelated reverb delay clusters adding like resonances inside a rectangular speaker cabinet, designed such that double ups are minimal. It reminds me of open baffle speakers in that it's technically a step away from absolute fidelity, but can give us something more enjoyable.
But if the delay times are that short as to not be perceived (2mS was mentioned), won't that cause the distance between the nulls to be perceptable by a person sitting near the L or R speaker? A sort of frozen phlanging effect? The article mentions that the improvements are only significantly perceptable in relatively dry environments. Maybe if this same technique was used with an actual center speaker, it might be a significant improvement. 3 different effectively decorrelated reverb delay clusters adding like resonances inside a rectangular speaker cabinet, designed such that double ups are minimal. It reminds me of open baffle speakers in that it's technically a step away from absolute fidelity, but can give us something more enjoyable.
Bob, I've tried it on a few occasions with my line arrays. It really did clear up the center sound, less dark. But for me it also pushed it a bit back. Plus the sound that's usually beyond my speakers left and right didn't go past the speaker position anymore.
The center image was more stable but I did hear a strange effect when moving to off axis. It stayed in the center longer. Without the phase shuffler the phantom follows me to the side I move to. With the shuffler it seemed to stay in the center a bit longer until I heard it a bit unclear as if I did hear it double. That was only the case with the first rough shuffler for me. The later one I tried was more subtle and did work more as advertised.
In the end I liked the mid/side S curve better(*) in my system with some adjustments. It does not disrupt the positioning and helps my depth perception. I already had the tools in line to do it and it had less draw backs for me. I kept that, but admittedly also because of ease as I already run convolution and couldn't get it to run twice without other problems.
(*) LoCo part of this post
Can't figure out why there haven't been more people stepping up to try this out. I certainly enjoyed reading the paper(s) and try it! Should be fun for movies too when you run a phantom center. Worth a try at least.
The center image was more stable but I did hear a strange effect when moving to off axis. It stayed in the center longer. Without the phase shuffler the phantom follows me to the side I move to. With the shuffler it seemed to stay in the center a bit longer until I heard it a bit unclear as if I did hear it double. That was only the case with the first rough shuffler for me. The later one I tried was more subtle and did work more as advertised.
In the end I liked the mid/side S curve better(*) in my system with some adjustments. It does not disrupt the positioning and helps my depth perception. I already had the tools in line to do it and it had less draw backs for me. I kept that, but admittedly also because of ease as I already run convolution and couldn't get it to run twice without other problems.
(*) LoCo part of this post
Can't figure out why there haven't been more people stepping up to try this out. I certainly enjoyed reading the paper(s) and try it! Should be fun for movies too when you run a phantom center. Worth a try at least.
Last edited:
I finally got my convolution fixed so I tried this out briefly. I also think that the shuffler pushes the image back a touch though I didn't notice any perceptible change in the image quality. I need to try it out some more. My speakers have a fairly narrow pattern from 300hz up, and my room is narrow/assymetric/untreated. It would be nice if there wasn't a click noise when you switch convolution on/off in JRiver for doing a quick A/B.
After pondering more, since I've already built a nice center speaker for under my 42 inch TV, I wonder if you would actually need to add the decorrelated reverbs to the L&R speakers when there is an actual center speaker? It seems like adding a recirculated 1 or 2 mS delay that dies out after 10mS, to only the center speaker signal, might be enough to create more of a sense of separateness and perceived clarity for the center image, and not reduce the perceived spaciousness of the L&R speaker signals that wesayso mentioned above... (?)
I also wonder if adding matched fixed non-circulated delays (maybe 10mS) to the L&R speaker signals would pull the center speaker image forward (?)
Sorry to be a bit off subject here. I know this thread is for those who don't want a center speaker.
I wish there was a simple to use hi-fi no noise 24 bit delay chip that would be perfect for this experiment (?) 16 bit? Maybe my Strymon delayer pedal could at least prove the theory good or bad.
I also wonder if adding matched fixed non-circulated delays (maybe 10mS) to the L&R speaker signals would pull the center speaker image forward (?)
Sorry to be a bit off subject here. I know this thread is for those who don't want a center speaker.
I wish there was a simple to use hi-fi no noise 24 bit delay chip that would be perfect for this experiment (?) 16 bit? Maybe my Strymon delayer pedal could at least prove the theory good or bad.
If you have a center channel, then no need for the phase shuffle. It's for stereo phantom image only.
If you are playing back from a computer, you can do the delay there in as many bits as you need.
Thanks for trying this guys, and reporting your results. Know what it does on different systems is important. There may not be a good "one size fits all" approach this. Hearing about what you hear helps a lot to know how well this works.
If you are playing back from a computer, you can do the delay there in as many bits as you need.
Thanks for trying this guys, and reporting your results. Know what it does on different systems is important. There may not be a good "one size fits all" approach this. Hearing about what you hear helps a lot to know how well this works.
Thanks for pointing me to this thread Pano!
Figure 1 is the Vickers paper is a physical representation of the notch depth based on summing two signal with the delay and head shadowing, but are you aware of any phychoacoustics representations? ie double blind studies where they sit many listeners in front of a stereo triangle in an anechoic chamber and then have them adjust the loudness of central image tones to match that coming from one loudspeaker only, then plot stats of the results?
Having worked on auralization a bit in my old job, I've read up on binaural addition, and I don't think the notches would be perceived to be quite as deep as the mathematical/physical summation (with head shadowing) would make them out to be, but that's just an (educated) guess.
Thanks for giving me something else to fret over
Figure 1 is the Vickers paper is a physical representation of the notch depth based on summing two signal with the delay and head shadowing, but are you aware of any phychoacoustics representations? ie double blind studies where they sit many listeners in front of a stereo triangle in an anechoic chamber and then have them adjust the loudness of central image tones to match that coming from one loudspeaker only, then plot stats of the results?
Having worked on auralization a bit in my old job, I've read up on binaural addition, and I don't think the notches would be perceived to be quite as deep as the mathematical/physical summation (with head shadowing) would make them out to be, but that's just an (educated) guess.
Thanks for giving me something else to fret over
I don't think the notches are as deep as the sims. But it is something I hear, for sure.
Usually if the overall balance is right, then sounds panned hard right or left sound too bright. If the middle dip is gone, then the overall tonal balance is a little too bight. I'd love to EQ just sounds panned hard to one side, but the shuffler fixes the center effect.
Usually if the overall balance is right, then sounds panned hard right or left sound too bright. If the middle dip is gone, then the overall tonal balance is a little too bight. I'd love to EQ just sounds panned hard to one side, but the shuffler fixes the center effect.
Reading the thread and what you're trying to do, it sounds allot like the old Carver Sonic Holography, or any speaker-based cross-feed cancellation. If you had access to etymotic in ear mics and could measure your own hrtf, that would be something!
Hi Pano - interesting thread.
I did want to point out that the above statement is incorrect as the two impulses do not add to zero - only alternate impulses cancel, the others are doubled. Hence there will be some frequency response changes resulting from this. Not sure to what extent and it may be a non-issue, but it is there.
It's different from crossfeed. It's a phase shuffle to eliminate comb filtering. Works, too!
I have measured my HRTF with DIY mics, but haven't gotten it to work well with headphones.
It's not needed, but its the ultimate of what you're trying to accomplish. You're obscuring the crosstalk impact through the shuffler, the crossfeed canceller attempts to cancel it outright.
IME working on this directly, you need your own measured hrtf to make it work without unacceptable tonal compromises
Could be, yes. But I made one of my own that is only a phase shuffle, no echos. It should sum basically flat, I think.
Not sure about that. If the comb filtering is cancelled by the shuffle, then no need for custom settings. Tho maybe the point at which it needs to start could be related to head size. I don't see it as crosstalk, but more as comb filtering. But of course if there were nothing common between the channels there should not be any comb filtering. Hmmm.....
It dawned on me that they should have used an echo with alternate phases of the impulses after the first. This would then cancel in the center and create a neutral frequency response while still shuffling the left and right channels. There would still be an aberration in the left and right channels.
As you say, a phase shuffler makes the most sense.
Maybe I haven't thought it through completely but it seems counter intuitive. Consider the HRTF directivity chart below. The ear 'expects' much less mid and high content from a source directly in front of the listener than one 45 degrees off axis. In a typical stereo set up, with speakers 45 degrees either side of the listener and equal content in both speakers to create a phantom centre, the listener will receive the mid and high energy of sources 45 degrees off axis but localization associated with 0 degrees. Assuming no compensation in production shouldn't it sound brighter instead of darker?
A useful experiment might be to compare level matched white noise played from one speaker then both to sort out if your experience is due to something happening upstream, i.e. the recording.
Attachments
Yes, I've tried it. I've even posted some test files in the thread - somewhere back there.
Here's the thing. If you have a real source straight ahead, it will sound normal, or at least how you expect it to sound. But if that straight ahead source isn't actually there, it's a phantom made by two speakers, then the tonal change happens. It happens because of the comb filtering between the two speakers.
Remember, it's only because of the stereo phantom image, you don't get the effect with actual sources located directly in front of you. They don't have the comb filtering.
Here's the thing. If you have a real source straight ahead, it will sound normal, or at least how you expect it to sound. But if that straight ahead source isn't actually there, it's a phantom made by two speakers, then the tonal change happens. It happens because of the comb filtering between the two speakers.
Remember, it's only because of the stereo phantom image, you don't get the effect with actual sources located directly in front of you. They don't have the comb filtering.