Hi Earl, I don't think it is a wild goose chase, but maybe we are talking two different things here. One is the audibility of phase and the other is the audibility of time alignment. Seemingly the same thing right? But I don't think it is. And hence why the confusion, at least it seems that way from the experiments I have done.
Have a look at this experiment on speaker alignment audibility: Some Experiments With Time
"Three experiments were performed which confirm the audibility of time offset in loudspeaker drivers but indicate that this audibility is due only to the frequency response aberrations resulting from the time offset. Implications of these results are discussed."
"Despite the possibility of "digital effects," the modest quality of the speakers, and the fact that the listeners were in effect being tested, every one of approximately 12 participants was able to match the time delays to within +/- 40 microseconds (about 1/2 inch). This finding confirms the clams of those who espouse the virtues of arrival time compensated loudspeakers."
This has been my experience as well, albeit my just noticeable difference is around 2 sample difference (48 kHz sample rate) at 500 Hz digital XO between a woofer and mid.
As I mentioned in post 284, the tone quality changes, for the better and this experiment seems to support that claim. It also might explain why folks that linearize the phase of their existing system, (but don't time align the drivers), don't hear any or minimal changes to the sound. It also might be that by careful time alignment, given the small values of time increments, one ends up with a flat phase response (and GD) as a by product, but suggesting that is the cause of the effect. I can't confirm the latter as I do use linear phase digital XO and excess phase correction, but I can independently remove the excess phase correction to see if I can hear an audible difference.
That's why I suggested in post 284 that a good experiment would be to vary each of these variables one at a time to determine the audibility of each one.
Regards, Mitch
Mitch - to me there is no difference between phase and time or more importantly group delay, which is derived from the phase. Arrival times is group delay and I have studied group delay before and can confirm that it is audible at some SPL level. And that aspect confounds this whole discussion because does tester A do this test at 70 dBSPL while tester B does it at 90 dBSPl, if so then the two results are not comparable. The audibility of group delay is SPL dependent!
You say "vary each of these variables one at a time to determine the audibility of each one" - specifically what are those variables?
I'd like to reiterate that Brian Moore (Prof. at Cambridge) studied group delay directly and found that it was audible over headphones but not in a "reverberant room" (whatever that means) over loudspeakers. So this is not a new question, but perhaps there are new aspects to it.
You say "vary each of these variables one at a time to determine the audibility of each one" - specifically what are those variables?
I'd like to reiterate that Brian Moore (Prof. at Cambridge) studied group delay directly and found that it was audible over headphones but not in a "reverberant room" (whatever that means) over loudspeakers. So this is not a new question, but perhaps there are new aspects to it.
Last edited:
Earl, not sure if you looked at the article... Some Experiments With Time
It was a double blind test that had 12 folks adjust the time delay between tweeter and woofer and all of them were able to dial it in to +- 40us of aligning the two drivers... There were two other experiments and the conclusion: "Three experiments were performed which confirm the audibility of time offset in loudspeaker drivers but indicate that this audibility is due only to the frequency response aberrations resulting from the time offset..." One of the experiments adjusted the phase, by a considerable amount, yet none of the participants could hear any difference... I think the article is worthy of read and discussion as that seems to imply phase linearization is inaudible, but time alignment is audible in the from of frequency response aberrations
I have another paper that had a similar experiment and the testers were able to hear down to a 10us offset. Yah, 10us. I will post it when I have some time later.
As noted in the article and maybe it is coming to light that it is the time alignment that folks are pointing to having the favorable sound and not necessarily the phase linearization...
It was a double blind test that had 12 folks adjust the time delay between tweeter and woofer and all of them were able to dial it in to +- 40us of aligning the two drivers... There were two other experiments and the conclusion: "Three experiments were performed which confirm the audibility of time offset in loudspeaker drivers but indicate that this audibility is due only to the frequency response aberrations resulting from the time offset..." One of the experiments adjusted the phase, by a considerable amount, yet none of the participants could hear any difference... I think the article is worthy of read and discussion as that seems to imply phase linearization is inaudible, but time alignment is audible in the from of frequency response aberrations
I have another paper that had a similar experiment and the testers were able to hear down to a 10us offset. Yah, 10us. I will post it when I have some time later.
As noted in the article and maybe it is coming to light that it is the time alignment that folks are pointing to having the favorable sound and not necessarily the phase linearization...
Last edited:
With no intentions of being critical of the work, I'm not sure the results are of any real value to the experiment if the intention was to determine the audible perceptions of time differences without all else 'being equal' where adjusting the FR of the test units would have been imperative to a relevant conclusion. The question of measuring to 40us is worthless when the actual measure is the simple decibel.
We know that physical offset inherently causes a phase shift as well as differing responses...........correct the phase and FR and retest........curious to see what comes of that.
We know that physical offset inherently causes a phase shift as well as differing responses...........correct the phase and FR and retest........curious to see what comes of that.
There was a follow up to this experiment where the delay was added before the speaker, which showed what you're suggesting that the response was what made it easier to detect.
Earl, an audio test I'd like to see run is more basic.
I've seen lots of claims on various audio forums where people's hearing tested "beyond 26k", "22k", "23.5k", I don't remember them all, but it isn't rare to see. (The assumption, I think is that being able to hear to higher frequencies implies you can hear more discriminately somehow?). Anyway, the stats I remember about this make these claims (or at least the number of them) seem kind of unlikely.
I wonder if what is being heard is really the tone or the sound of something else reacting (boiling voice coils, aliasing in a sound card, or something similar). So a do-it-yourself test I'd like to see is a hearing test that tests more than just hearing "something" when so stimulated. A set of recorded tests with tones proceeding up the scale octave-by-octave, but the highest one (the frequency being tested) being either the right octave tone or a really sour nearby tone. Maybe a series of those kind of tones, listener picks the one that is on-key. Because it probably doesn't matter if you can hear that high if you can't tell if it is a good musical response..
Perhaps something equivalent to this is already done, though in the few hearing tests I've had done, it was simple "push a button when you think you hear something". Those tests also went nowhere near as high as 20kHz, too, maybe because of the subjects age!
I've seen lots of claims on various audio forums where people's hearing tested "beyond 26k", "22k", "23.5k", I don't remember them all, but it isn't rare to see. (The assumption, I think is that being able to hear to higher frequencies implies you can hear more discriminately somehow?). Anyway, the stats I remember about this make these claims (or at least the number of them) seem kind of unlikely.
I wonder if what is being heard is really the tone or the sound of something else reacting (boiling voice coils, aliasing in a sound card, or something similar). So a do-it-yourself test I'd like to see is a hearing test that tests more than just hearing "something" when so stimulated. A set of recorded tests with tones proceeding up the scale octave-by-octave, but the highest one (the frequency being tested) being either the right octave tone or a really sour nearby tone. Maybe a series of those kind of tones, listener picks the one that is on-key. Because it probably doesn't matter if you can hear that high if you can't tell if it is a good musical response..
Perhaps something equivalent to this is already done, though in the few hearing tests I've had done, it was simple "push a button when you think you hear something". Those tests also went nowhere near as high as 20kHz, too, maybe because of the subjects age!
Bill - excellently thought out experiment, but here is the problem. Lidia would laugh if I told her I wanted to test people who claim to hear above 20 kHz. I could not get her to take it serious.
Phase and/or group delay detection has significant implications to what she does. Just like Brian Moore's study of non-minimum phase in a loudspeaker - it's way out of his league (below it!) But I'll bet the money was right. Still he had to see some connection to his real job. Lidia though that the greater detection of group delay with SPL was quite interesting.
Phase and/or group delay detection has significant implications to what she does. Just like Brian Moore's study of non-minimum phase in a loudspeaker - it's way out of his league (below it!) But I'll bet the money was right. Still he had to see some connection to his real job. Lidia though that the greater detection of group delay with SPL was quite interesting.
Last edited:
Mitch - just for you, I'll read it.
I still see time alignment and phase linearization as the same thing. If there is flat phase across the bandwidth then the excess group delay must be zero and hence the time alignment has to be perfect. They are the same thing!
Still, I want to know what the variables are as you see it - physical speaker offset? That takes way to much construction to do.
Am I missing something or are your tests contradicting each other. One same that timing does not matter and the other says it does. Maybe I've just had too many drinks.
I saw the David Clark test when they were first done. I am in the SMWTMS group when he did that.
When I built my first generator I showed I could hear 24-28kHz IIRC. I was using a Vifa dome, the only control was that I swept up to confirm I could follow the tone, and I think it's plausible.
Earl, thanks for the input. I've wondered how a real hearing pro would respond to such claims. Do you happen to know what the stats (pecentage) are for, say, people over 30 able to hear to 20kHz at only a few dB down?
When I see those kinds of claims, I do kind of imagine that the poster must be sitting in a parent's lap having mom type it in, likely not old enough to read or write yet!
When I see those kinds of claims, I do kind of imagine that the poster must be sitting in a parent's lap having mom type it in, likely not old enough to read or write yet!
Earl I could help with the test if you want to use something like rePhase and Foobar ABX. You'd need a HOLM sweep of your speakers to see what phase is doing. Or it could be estimated from box and crossover. rePhase can correct phase only, no change in amplitude. A correction impulse is generated that is used in the playback convolution engine. Or the convolution can be done offline and applied to the wave file. With Foobar ABX you can smoothly switch between the normal and corrected phase files to test audibility.
Should not be a difficult test to do.
I find the effect subtle, but noticeable. Not sure if I like it or not. That might be a long term decision, made after weeks of listening.
Should not be a difficult test to do.
I find the effect subtle, but noticeable. Not sure if I like it or not. That might be a long term decision, made after weeks of listening.
Earl, Merry Xmas!
Off the top of my head…take a 2 way speaker and design a linear phase digital XO implemented as a FIR filter as we want to adjust amplitude and phase independently. While I use Acourate, my limited understanding is that rePhase should be able to this. Choose a target frequency response as we want both test setups to have the same frequency response. The single unit under test here is the time alignment (i.e. digital delay) as follows:
Test 1 FIR – delay the woofer XO by some arbitrary amount like 1 millisecond for example. Sound travels roughly 1 foot per ms, so in the 2 way that puts the acoustic center of the woofer about a foot behind that of the tweeter relatively speaking. I can do this in Acourate, but not sure if RePhase can do this. Then linearize amplitude (to target fr) and phase, generate FIR filter.
Test 2 FIR – identical to Test 1, except the digital delay is set so that the woofer and tweeter are exactly time aligned. Again linearize fr and phase. Generate the 2nd FIR filter.
Using JRiver or other Convolver, load Test 1 FIR filter and listen to your favorite tunes. In JRiver, one can switch FIR filters on the fly while music is playing with a small (silence inserted) interruption.
My claim is that there should be a) an audible difference between the 2 filters and b) the time aligned filter will sound better than the delayed bass filter 🙂 Even though both amplitude and phase have been linearized.
I can do this in Acourate where I can correct both amplitude and excess phase, but can also independently adjust the delay (by the number of samples) in each leg of the XO like this:
I then use the same target frequency response and excess phase correction settings in both filters with the only change being the delay. Not sure if this can be done in rePhase. While in Acourate, I can apply the delay before or after amplitude and excess phase correction, for the purpose of this test, the delay should be applied first and then linearize amplitude and phase.
Of course there are several things to keep track of, ensuring the FIR filters are level matched, etc.
Thoughts?
PS. Nate, Merry Xmas to you and sorry for going off topic. Hopefully some of this will be useful for your next iteration of your Synergy build down the line.
Not sure about this - surely you are only correcting the phase and amplitude in the region of the measurements, and therefore a 1ms time delay on the woofer would have a much wider, unmeasured effect on power response, lobing, and frequency response elsewhere.
I may be wrong - but surely if you took measurements horizontally and vertically in 10 degree increments off axis of both your test set-ups they would look very different in both phase and frequency response (even though the same in the mic position that had been corrected by the convolver).
It would therefore be obvious that they would sound very different?
I may be wrong - but surely if you took measurements horizontally and vertically in 10 degree increments off axis of both your test set-ups they would look very different in both phase and frequency response (even though the same in the mic position that had been corrected by the convolver).
It would therefore be obvious that they would sound very different?
Mitch - first let me thank you for opening my eyes to something that I had missed in this discussion and for a long time. I kept thinking about this and then after several Bourbons, its 1 am and I am listening to music and it hit me:
If one tries to correct a loudspeaker as a system, i.e. with a single input to the system, then only the pressure response can be corrected, the power response is fixed by the relationship between the drivers and no amount of phase/group delay change can have any effect on the power response. However, if one put the EQ in line with each individual driver then one can not only affect the pressure response but the power response as well - one can therefor correct both at the same time. This is easy to see if you think about it.
I am still thinking through the implications of this epiphany, but it is clear that what I have said for a long time - that electronics cannot correct a loudspeaker "system" is absolutely true. But electronics can improve the system if it is implemented on a per driver basis. Another implication is that if one has a system with a flat frequency response and a flat power response then there is no phase (or group delay) adjustment that can improve the situation - its already optimized. This is completely consistent with what Toole/Olive have been saying all along - if you have a correctly designed loudspeaker system then there is no room for further correction with electronics and phase correction of the "system" will not improve the situation, i.e. it is not worth studying.
I don't know if this is what you were saying, but in trying to understand what you were saying this concept dawned on me.
PS. This is exactly what I have been doing in my speakers all along. The drivers were always time aligned and once I had a flat listening axis response with a flat power response, i.e. a flat DI, then there isn't any room for more correction. I did find that small further corrections could be made with a bi-amp situation, but they were not pronounced since the system was already nearly optimized. So in your test there would clearly be an audible difference. Proving that would not be very interesting I don't think.
If one tries to correct a loudspeaker as a system, i.e. with a single input to the system, then only the pressure response can be corrected, the power response is fixed by the relationship between the drivers and no amount of phase/group delay change can have any effect on the power response. However, if one put the EQ in line with each individual driver then one can not only affect the pressure response but the power response as well - one can therefor correct both at the same time. This is easy to see if you think about it.
I am still thinking through the implications of this epiphany, but it is clear that what I have said for a long time - that electronics cannot correct a loudspeaker "system" is absolutely true. But electronics can improve the system if it is implemented on a per driver basis. Another implication is that if one has a system with a flat frequency response and a flat power response then there is no phase (or group delay) adjustment that can improve the situation - its already optimized. This is completely consistent with what Toole/Olive have been saying all along - if you have a correctly designed loudspeaker system then there is no room for further correction with electronics and phase correction of the "system" will not improve the situation, i.e. it is not worth studying.
I don't know if this is what you were saying, but in trying to understand what you were saying this concept dawned on me.
PS. This is exactly what I have been doing in my speakers all along. The drivers were always time aligned and once I had a flat listening axis response with a flat power response, i.e. a flat DI, then there isn't any room for more correction. I did find that small further corrections could be made with a bi-amp situation, but they were not pronounced since the system was already nearly optimized. So in your test there would clearly be an audible difference. Proving that would not be very interesting I don't think.
Last edited:
bushmeister, if what you are describing/thinking is true it would make an excellent argument to test this on speakers, not headphones to get the complete picture.
I can't exactly abx with my setup (convolution has too much delay while switching) but the graphs I showed earlier, one with linear phase and one minimum phase do sound different to me while having the exact same FR graph. Even though the differences between them are small when viewing the group delay plot.
Linear phase group delay plot:
Minimum phase group delay plot:
While this seems like a really small difference the second one (minimum phase) sounds way more natural to me. No "pok" or "pak" differences between them, just differences in how I perceive the bass signals.
I can't exactly abx with my setup (convolution has too much delay while switching) but the graphs I showed earlier, one with linear phase and one minimum phase do sound different to me while having the exact same FR graph. Even though the differences between them are small when viewing the group delay plot.
Linear phase group delay plot:
Minimum phase group delay plot:
While this seems like a really small difference the second one (minimum phase) sounds way more natural to me. No "pok" or "pak" differences between them, just differences in how I perceive the bass signals.
Last edited:
Member
Joined 2009
Paid Member
If time alignment is the same thing as preserving the signal waveform then that's what I want. Never understood how multi ways can work properly just based on dividing up the signal into frequency ranges and hoping it all comes together in the brain. I want the signal waveform out of the source to be preserved by the speaker.
Yep wesayso - I may have to install some software to try this on mine - I run with minidsp 4x10hd, so no phase correction for me at present.
I would love to see a full set of off axis measurements with and without phase correction.
I would love to see a full set of off axis measurements with and without phase correction.
bushmeister - Maybe my choice to go with a line array with full range drivers becomes makes more sense now to try this all, compared to a multi way setup. If I move up or down or left to right compared to my speaker I'm not changing any relative distances to any driver firing only a part of the frequency spectrum. in other words it doesn't change the relationship like it would with a woofer and a tweeter on a baffle.
I can't easily measure a polar in my room with these towers, I'll leave that up to you guys to test. I did test what was of interest to me as described earlier.
I can't easily measure a polar in my room with these towers, I'll leave that up to you guys to test. I did test what was of interest to me as described earlier.
Could you show an impulse, STEP and group delay plot from a measurement taken at the listening position?
To me this still sounds like you accept the phase rotation due to the crossover type(s) used. Resulting in group delay.
With the separate driver control in Acourate you could have the same needed slopes without the phase rotation. So basically without the group delay.