Speaker time alignment, does it matter?

[Inner Space]

I'm sick in bed ...

Nothing sinister, I hope, Ben. Best wishes and get well soon.

 

[BenB]

There was an investigation already, cited by somebody else, peer reviewd, to some degree by Toole. What are You aiming for?

It appears that the literature isn't entirely convincing to a lot of people. When audible differences are perceived on synthetic waveforms, I understand when people wonder whether it might make an important difference to THEM on musical material. We also have literature (sales material and interviews) from experts in the field claiming phase distortion is basically the reason speakers don't sound like live instruments (or at least a big part of that).

My aim is to put the tools in people's hands to run the test themselves and see just how subtle a difference we are talking about here. It's obvious from the anecdotes that hearing is believing. If that doesn't happen, then keine sorgen, I didn't spend much time making them.

 

[KSTR]

It would be great to have another contributor check these for correctness.

Looks completey OK to me. Phase display for the 3kHz looks a bit distorted by the underlying sub-sample shift you have used but of course group delay shape is full spot on, the small base offset of -15us is irrelevant (as it is constant). For ABXing this in foobar, a dummy kernel would be required with the same length and the dirac at the same position to avoid clues from time offsets and/or convolution artifacts etc. Also, crossfade shouldn't be used during ABX, for the same reason (crossfade is considered illegal for official ABX anyway, but I find the hard switching distracting, a fade-out + fade-in would be much better).
Most convolvers will take 32-bit floating point data, btw.

IME the 100Hz should be audible for most people with some specific music material, both as a slight timbre shift of steady-state bass notes (when having strong 2nd harmonic) as well as the slight time lag of the "body" of kick drums and plucked upright bass notes. With linear-phase speakers the differences are more pronounced and of course strong room modes can be a show stopper.
Headphones also would apply but for reasons I don't fully understand atm I find phase issues harder to detect with headphones than with speakers.

 

[BenB]

Looks completey OK to me. Phase display for the 3kHz looks a bit distorted by the underlying sub-sample shift you have used but of course group delay shape is full spot on, the small base offset of -15us is irrelevant (as it is constant). For ABXing this in foobar, a dummy kernel would be required with the same length and the dirac at the same position to avoid clues from time offsets and/or convolution artifacts etc. Also, crossfade shouldn't be used during ABX, for the same reason (crossfade is considered illegal for official ABX anyway, but I find the hard switching distracting, a fade-out + fade-in would be much better).
Most convolvers will take 32-bit floating point data, btw.

IME the 100Hz should be audible for most people with some specific music material, both as a slight timbre shift of steady-state bass notes (when having strong 2nd harmonic) as well as the slight time lag of the "body" of kick drums and plucked upright bass notes. With linear-phase speakers the differences are more pronounced and of course strong room modes can be a show stopper.
Headphones also would apply but for reasons I don't fully understand atm I find phase issues harder to detect with headphones than with speakers.

Wow, that didn't take long. Thank you very much for doing that. I also appreciate the information you have provided.
If you looked at the filters themselves (in the time domain) you should notice that the dirac is right at the first sample (no pre-ringing here!). With that in mind, I had assumed that no dummy kernels would be required. Let me know if you disagree. I'd like to get the information together so it's as easy as possible for people to test this themselves, and it sounds like you have more recent experience with this than I do.
That's good information that the convolvers can take floats. I noticed that by default foobar wanted a file in a music format like .wav, so that's what I made. The nice thing about that is it makes it obvious where to cut the filter off. When the response rounds down to 0 in 16 bits, that's the end.

I am posting an example of the filter, the magnitude, and phase response from my own analysis. Please let me know if you see things any differently.
Note that the scale on the magnitude response will be +/- 2 thousands of a dB, so it will look like a mess when in reality it's basically perfectly flat.

Regarding the audibility, I believe your statements are mostly consistent with my recollections, with a few exceptions. Would you care to comment in preference? Also, do you think I should add an even lower crossover frequency to the mix? I don't think anything less than 80 would be common.

 

[KSTR]

Let me know if you disagree.

As mentioned, for proper ABXing with foobar's ABX pluging. Convolver on vs off makes a timing difference that would have to be adjusted manually otherwise (render to wav's and trim them manually with a high risk of getting it wrong unless you put in a marker pulse). And IME it is prudent to change only one single variable in a compare, which is the kernel in this case, and leave everthing else as is. We're not seeking for false positives here.

WAV is a container format and supports a full variety of stream formats, not even restricted to simple linear PCM.

The plots are identical to what is seen when loading the IR into REW.

Signals with frequencies below 80 or 100Hz also give information, mostly on the timbre side of things but are not essential IMHO. Timing-wise we run into the phase rotation of the system high-pass (minphase) which adds to the allpass phase.

Not sure what you mean with preference. What I can say is that the timbre shift aspect is pretty much arbitrary wrt what is preferred as this is dependant on the phase of the harmonics in the recording and the system's phase response. The timbre difference of strong 2nd harmonic is strongest when the phase offsets, for the ear signal, are 0°and 180° (with cosine generators, of course, else its 90° and 270°) and for that reason then also a simple poarity switch has the same timbre shift effect. Whereas, with 90° vs 270° phase of 2nd there is no change as the waveform is the same in both cases (only "inverted in time" shape which is irrelevant for the spectral distribution we hear).

The timing thing of course is clear, I always prefer the faster timing. I even went to partly unwrap the system roll-off which makes a worthwile improvement if you have, say, a 60Hz ported speaker active speaker with 2nd order "infransonic" filter, making a 6th order phase response at a rather high frequency. If you reduce the phase to 3rd or 2nd order there is an improvement in "speed and coherence with the rest of the signal".... but also a worsening of "pulse compactness". This is unavoidable as the system roll-off isn't minphase anymore and must contain "pre-ringing". It is not a true ringing and doesn't sound like one either, rather the time-inversed sort-of exponential decay (hence, attack) sounds more like a noise signal increasing in volume, which is seldom annoying with typical music signals although with step exitation (or 0.5Hz squares etc) it can be heard.

 

[QMuse]

I'm hardly going to shut down the discussion or ban all talk of it as you so bizarrely suggest .

Where in my posts did I suggest such thing? The only thing I did suggest is to our happy camper to search the forum for the links and quotes from the previous such discussions.

Now there's people here discussing time alignment, not for the first time and likely not the last.

Yes, there are, and I am one of them, sharing knowledge, experience and measurements. But he is not discussing anything nor he is sharing anything except his personal claim that what he heard god knows how many years ago sounded good because the speakers were time aligned. He also expressed very clearly that he's not willing to accept scientific arguments, so tell me what exactly is he doing here?

 

[QMuse]

This is unavoidable as the system roll-off isn't minphase anymore and must contain "pre-ringing". It is not a true ringing and doesn't sound like one either, rather the time-inversed sort-of exponential decay (hence, attack) sounds more like a noise signal increasing in volume, which is seldom annoying with typical music signals although with step exitation (or 0.5Hz squares etc) it can be heard.

Regarding pre-ringing and how it sounds - here is a short music sample which contains 4 drum kicks without pre-ringing and same 4 drum kicks with pre-ringing.

 

[KSTR]

^ Filter specs?

 

[GelbeMusik]

It appears that the literature isn't entirely convincing to a lot of people. ... might make an important difference to THEM … claiming phase distortion is basically the reason speakers don't sound like live instruments (or at least a big part of that).

My aim is ...

If people do not trust the literature, I confirm that such is a healthy attitude. But only so, if they bear the burden to prove the literature wrong themselves. As I stated before, most regularly there is a conflict, though. On one hand the literature is not accepted, on the other hand any competence and / or motivation to actually work on the topic is lacking.

I would go even further, as I stated above. There is a desire to not (!) trust the stereo system. To always scrutinize its realism, to doubt it being true to the source. I think that the expectations behind this doubt are maybe the worst misunderstanding in all audio.

Actually:
- stereo is a basically flawed concept, it doesn't work theoretically and practically (Toole)
- there will never be possibility to transfer the bare signal of an auditory scene from the origin to a record to the ears (Toole)
- hence the relevant information has to be identified (Toole)
- as being frequency response in-room and volume and non-linear distortion (Toole)
- done by widest margins (Toole)

If some folks think, they should be served better than all others on the world, and so they select their poison as being "time", this demand can be satisfied. They only have to pay! Yes, not every extraordinary wish can be fulfilled for free, taking others into obligation--You. There is "Dirac" to correct "time", "phase" whatever they name their extra extra. And other possibilities.

Of course they would need help. I personally would refuse it. To doubt all the literature, and then not being able to setup "Dirac"? To request extraordinary service with super special properties of their very personal stereo, and then not willing to pay for this extra extra service, giving something back, basically?!

C'mon!

 

[QMuse]

^ Filter specs?

Last 4 kicks were created convolving a room EQ filter with excessive phase correction which contained several high Q phase filters into first 4 kicks. Filter had very visible pre-ringing in step response but very flat excess phase and GD. The point was to demonstrate that it makes no sense to EQ like that as you make more harm than good because pre-ringing is audible while GD isn't.

 

[Thomas savage]

Where in my posts did I suggest such thing?

Here ..,


" No problem, I'll see my self out, as it seems on this forum moderating doesn't seem to imply cutting off the futile discussions that happened many times before, but in the name of "democracy", it is about letting as many new users to repeat the same bullshit all over again. "

Forums are full of individuals at different stages of learning and levels of understanding, things will get repeated .

If you don't like it , don't participate . Certainly don't start insulting me with accusations of profiteering and incompetence. That will lead you to the exit door, you have had your one free pass .

 

[QMuse]

Here ..,


" No problem, I'll see my self out, as it seems on this forum moderating doesn't seem to imply cutting off the futile discussions that happened many times before, but in the name of "democracy", it is about letting as many new users to repeat the same bullshit all over again. "

Forums are full of individuals at different stages of learning and levels of understanding, things will get repeated .

If you don't like it , don't participate . Certainly don't start insulting me with accusations of profiteering and incompetence. That will lead you to the exit door, you have had your one free pass .

So, you were insulted by my opinion that moderating a forum isn't the same thing as policing a forum? IMO it is moderator task to warn users to use search button instead of asking for same quotes and links to the articles that have been quoted/linked for at least dozen of times, but ok, it is up to you to choose how to do your job here.

 

[Thomas savage]

So, you were insulted by my opinion that moderating a forum isn't the same thing as policing a forum? IMO it is moderator task to warn users to use search button instead of asking for same quotes and links to the articles that have been quoted/linked for at least dozen of times, but ok, it is up to you to choose how to do your job here.

Please read this ..,
https://www.audiosciencereview.com/...-update-new-and-old-members-please-read.2596/

I'm issuing you with a thread ban , these public forum threads are not for members to debate forum management decisions or speculate on their motives . You have explicitly expressed a desire not to continue to participate in this thread , I will take away that temptation for you.

 

[BenB]

As mentioned, for proper ABXing with foobar's ABX pluging. Convolver on vs off makes a timing difference that would have to be adjusted manually otherwise (render to wav's and trim them manually with a high risk of getting it wrong unless you put in a marker pulse). And IME it is prudent to change only one single variable in a compare, which is the kernel in this case, and leave everthing else as is. We're not seeking for false positives here.

I think before anyone considers doing actual ABX testing, they would want to do some exploration about the impact of the all-pass filters on different songs or sounds. There's no point in setting up an ABX test for material that isn't right to highlight the difference. I made a DSP Chain Preset for each of the all-pass filters I posted, and also a preset for "none". After right clicking the toolbox and activating the DSP switcher, I can switch between the filters (and no filter) in realtime while listening to anything I want. Honestly, I think that would be enough for most people to realize how subtle of an issue this is. I didn't notice any annoying time jumps when switching between my filters, nor when selecting "none".

 

[dasdoing]

I would suggest finding personal threshold of delay hearing with this first https://www.audiocheck.net/blindtests_timing_2w.php?time=20
20ms should be easy. if it is, go down to 10ms and so forth.
If you end up not beeing able to hear 10ms (example) it would make no sense to correct 10ms of missalignment

 

[Snarfie]

To conduct a proper test of this hypothesis, it would be necessary to build two speakers that are identical in every respect except that in one, phase undergoes rotation from low to high speaker in a manner like the graph above for the KH310, while in the other, there is very little phase rotation, ideally none at all. This would be a very difficult thing to do, and as such, there may never have been a proper test of this question. Perhaps with a single-driver speaker, listened to close up, digitally equalized to achieve a near-flat response, one without any phase rotation intentionally introduced, the other with phase rotation introduced to mimic speakers like the KH310. Is it possible to do something like this with DSP?

What i did is far from what you propose but never the less.. I use Mathaudio Room EQ it Corrects both amplitude and phase components of frequency response and much more https://mathaudio.com/room-eq.htm . I tested / did measurements with several column loudspeaker ancient brands from more ore less the same depth en hight approx a meter high all bass reflex system from which 2 with a symmetrical load (build in woofer). Brands where Elipson 1303, JK Acoustics Optima 3 an Vandersteen model 1. Further more i i have some listening experience with Quad ESL 63 an IMF TDL speakers.

When i compared the speakers after measurements/correction the Vandersteen was clearly the speaker with the best imaging/transparent sound basically it was like a veil was lifted compared to the other speakers. Do know i listen for almost 35 years to a good quality speaker like the JK Acoustic Optima 3 which i like very much till i heard the Vandersteen revealing more information like the reverberation of studio 2 of the Beatles - Abbey Road recording i never heard before (difficult to explain they sounded suddenly more like a band more layers/space between instruments voices more intimate). I know this not an scientific approach an probably highly subjective but if speakers are corrected for phase an time delays as Mathaudio does did we not create more or less a level playing field for all speakers regarding at least a flat frequency response for all an taking out horrible room acoustics?

The speakers above are quite ancient so I planning to listen/test the Elac Debut Reference DBR-62 Amirm tested in combination with a proper sub woofer (if needed).

 

[BenB]

I think before anyone considers doing actual ABX testing, they would want to do some exploration about the impact of the all-pass filters on different songs or sounds. There's no point in setting up an ABX test for material that isn't right to highlight the difference. I made a DSP Chain Preset for each of the all-pass filters I posted, and also a preset for "none". After right clicking the toolbox and activating the DSP switcher, I can switch between the filters (and no filter) in realtime while listening to anything I want. Honestly, I think that would be enough for most people to realize how subtle of an issue this is. I didn't notice any annoying time jumps when switching between my filters, nor when selecting "none".

I eventually took the time to make some sample files to support ABX testing of phase distortion associated with typical crossovers. The thread is in the psychoacoustics forum:
https://www.audiosciencereview.com/forum/index.php?threads/phase-distortion-abx-testing.18709/

I fully expect complaints that I picked the "wrong" material, but I picked what I picked. I have provided impulse files in this thread and that thread so people can make their own samples.

 

[David A. Young]

Back in the 1980s, time-aligned loudspeakers became fashionable, such as the KEF105, KEF107, B&W DM6 (pregnant penguin) etc etc. I did a few experiments with a movable tweeter on a loudspeaker that could be moved back and forth by some 20cm/8". The idea was to hear whether there was any audible effect from moving the tweeter from behind the woofer, so late, to in-line, to forward so early.

Sadly, I and my colleagues couldn't hear any difference over that sort of distance, which was already rather more than any time misalignment due to drivers being on the same baffle.

I can imagine that if the drivers were so far apart that the time of arrival was very different, say a few seconds, then clearly there would be a mismatch between low and high frequencies. So, if a few seconds is too long, what about shorter times? Don't know at what point the two sounds will be perceived as co-incident, but certainly, over the few hundred microseconds of lead/lag I experimented with, there was no audible difference.

With my own current B&W 801s, I used the time-alignment facility on my DCX2496 crossover, just because it's there, but can't say again any difference was audible before and after.

Given the likely path differences between drivers on a typical loudspeaker, I don't see it as significant
S.

I recently found a pair of Paisley Research Model 20's at a local Value Village. These are Canadian made speakers from 1983-84 and have an 8" woofer above a 1" dome tweeter that is mounted on a panel behind the 1" thick baffle. The resulting cavity houses a foam and fiber-glass plug that surrounds the tweeter. The time alignment is clearly audible. Sitting 4' from the left speaker, you can hear a coherent wavefront. If the tweeter was on the same plane as the woofer, you would hear things shift between the two drivers. I also own various Tannoy co-axials ranging from 6.5 to 15 inch models. There are Technics SB-7000A's in my workspace and a pair of Energy 22 Reference Connoisseurs. All incorporate time alignment with audible results. I do have normal 2-ways from Rega-Camber, Paradigm, and ESM as well, but they aren't as nice to listen to when compared to something that has been time aligned.

One of the CD's played recently was a Denver Symphony Pops recording of Geoge Gershwin plays Rhapsody in Blue conducted by Newton Wayland. It sounds like it was recorded using a stereo overhead microphone setup. You can hear the room reverb clearly and the depth of the orchestra. This was released in 1988.

I used to build stage monitors and could never get horns that were deeper than the woofers to focus properly. The solution was to use a copy of the horn Altec used in the Model 14. The first 2 way loudspeaker to implement time alignment was the Altec A-7 "Voice of the Theatre" That design came into being because the folded LF horn/multicell HF horn combination of the day could not reproduce castanets properly because of the time delay caused by difference in horn lengths.

 

[David A. Young]

If your speaker baffle is not sloped like Thiel, Avalon etc... then you can raise the front part a little bit, make them sloped and listen. Simple experiment. If you hear a change in soundstage (messed up since it isn't designed for that) then time alignment might matter. Not really proof, but merits more research...

I repair loudspeakers and recently had a pair of EPI 200-C's in. During the listening tests, 3 things were noticed. 1-They were bass heavy. 2-The tweeter was not at the right height. 3-The tweeter didn't integrate well with the woofer. The solution was to remove the 1.5" high risers and replace them with 10" high sloped stands. This solved all three problems and made them listenable.

 

[David A. Young]

I am of the opinion that correct time alignment in a loudspeaker system does matter. I own a pair of Technics SB-7000A's, various Tannoy's and a pair of Paisley Research Model 20's. Ian Paisley solved the problem by using a 1" thick baffle and mounting the tweeter to a board glued to the rear of it. He also made a foam and fiberglass plug to fill the space between cavity walls and the tweeter's face plate. He also mounted the woofer above the tweeter and tweaked the crossover. The net result is a sense of the midrange being in focus and a noticeable sense of depth on recordings done with minimal microphone setups in a concert hall. I have also noticed the effect when lining up a 2" compression driver on a Emilar 820 horn with a 12" woofer. With the horn about an 1-1/2" forward of the woofer, the sound was more coherent. When I was designing and building stage monitors, I found that using horns that were deeper than the woofer never sounding very good and reversing the phase didn't make much difference. Using a short horn did.