The results with headphones will never reach the "fight or flight" reaction as a speaker can mimic. That's what I'm interested in, how we hear and feel the music and experience something that can trick us into thinking we hear something "real". I.m.o. it would be far fetching to come to any conclusions on audibility of phase based on headphone listening alone. Get a flat phase result at the listening position for a more honest review. That would take far more effort but would also be far more fair as judgement on this subject.
That talk is very interesting. Makes me think that I should use some test signals consisting of spoken words, speech. Griesinger showed some waveforms of the word "one" that consisted of an impulse train, and that could really highlight group delay differences if his description of hearing is correct.
Last edited:
What I want to know is: for a loudspeaker that uses certain crossovers, does the group delay from the crossover cause any audible effects?
I cannot use a loudspeaker to answer this question, because each loudspeaker will have its own group delay response and the imposed group delay in the test signal will add on top of that. With loudspeakers, when some audibility threshold is found, how would I know how to extract the influence of the GD in the imposed signal? The loudspeaker itself might already have a GD response that is close to the audibility limit.
OTOH with headphones, the data I have seen shows that GD is relatively "flat" (phase is relatively linear) for many models, at least much more so than any loudspeaker. This seems to be to be a much better test platform for my needs.
Well, to test for group delay with loudspeakers which IMHO is much to prefer vs. headphones for the more subtle effects like timbre and sound-stage, the speakers must not have any excess phase, that is the group delay / phase has to be minimum-phase.
In this day and age that's not too hard a task for a given multi-way speaker if your playback is computer-based and once you have the phase correction kernel to feed a convolver, it is easy to switch it on/off (off with dead-time correction, of course when you want to use it for ABX). One of the simpler tests is to use the correction kernel time-inversed which applies the excess phase twice and look if you find that audible or not.
My take is you are correct Charlie.
I think we need a speaker with zero group delay to test with; I don't think we can start with a speaker that already has its own group delay baked in.
So I believe a speaker with flat phase at zero is the proper stating test bed.
Whether that flat phase at zero is achieved via linear-phase or 1st order IIR xovers doesn't matter I think. Flat phase is the starting test bed goal.
Btw, very nice project you've undertaken.....thank you. It should evoke some thoughtful conversations.
I'll be surprised though, if any firm consensus develops with speakers. Between a room's masking, stereo masking (for those not testing a single speaker mono), and the issue of prebaked group delay, i think there is just too much noise going on.
I've played with speakers that do have flat phase at zero, other than the very bottom end roll-off.
In substituting IIR xovers for linear-phase xovers, so far I'm led to believe the lower the frequency of the crossover, the more I confident i get it's audible.
If the IIR order is raised beyond 4th, and xover frequency is low, I get pretty sure it's audible.
Once in the upper half of the spectrum, say above 640Hz, I mainly shrug I dunno.
I don't know where you have placed the group delay knees frequency wise, but if none are down low, you might want to try that.
I'm convinced the knees, where envelope distortion is created are the issue. I'm convinced the "relative" group delay between the high-pass side, and low pass side is not real time at all, and has no audibility unless folks tried to use it as a constant delay for time alignment. In which case a mistake was made, and may well be easily audible in comparison to correct time alignment.
my2c
@mark100 There is a nice paper (AES Convention paper) that hints at your own observations in its conclusions. It's available online:
https://acris.aalto.fi/ws/portalfil...udspeaker_Group_Delay_Characteristics_AAM.pdf
The group delay profiles for 3-way systems is what I would like to test next. Mostly for the 2-way systems (having only a single crossover with Fc above 1k Hz) the GD was not typically detected, but when the DG was elevated in a 3-way system by another crossover in the several-hundred-Hz region things were more detectable. I'm really not sure why the testing was done with "impulse responses and their time-reversed versions"... just compare an unprocessed signal with the one processed a la a 3-way speaker and use headphones. Anyway, that is where I am heading next. I have not exactly figured out how to adulterate the signals... make sure something is audible, or only represent "real world" loudspeaker group delay profiles. Will have to think about the experimental design a bit.
P.s. Some related info here, from a whitepaper by Scott Hinson:
https://www.audiosciencereview.com/forum/index.php?attachments/group-delay-pdf.31763/
I think he never got around to Part II...
https://acris.aalto.fi/ws/portalfil...udspeaker_Group_Delay_Characteristics_AAM.pdf
The group delay profiles for 3-way systems is what I would like to test next. Mostly for the 2-way systems (having only a single crossover with Fc above 1k Hz) the GD was not typically detected, but when the DG was elevated in a 3-way system by another crossover in the several-hundred-Hz region things were more detectable. I'm really not sure why the testing was done with "impulse responses and their time-reversed versions"... just compare an unprocessed signal with the one processed a la a 3-way speaker and use headphones. Anyway, that is where I am heading next. I have not exactly figured out how to adulterate the signals... make sure something is audible, or only represent "real world" loudspeaker group delay profiles. Will have to think about the experimental design a bit.
P.s. Some related info here, from a whitepaper by Scott Hinson:
https://www.audiosciencereview.com/forum/index.php?attachments/group-delay-pdf.31763/
I think he never got around to Part II...
I always find the discussion of group delay makes my head spin. The term 'group delay' seems to mean almost nothing itself. As I understand, it is only the rate at which the phase response is changing. This says nothing of the cause, which presumably could be anything from an all-pass filter to diffraction, cabinet resonances or the minimum-phase behaviour of an LC filter.
Would anyone agree the cause of the group delay is possibly more important to audability than the group delay value itself? Or am I way off??
EDIT: Still I'd like to thank and congratulate @CharlieLaub for going the extra distance and making a test rather than just talking like me!
Would anyone agree the cause of the group delay is possibly more important to audability than the group delay value itself? Or am I way off??
EDIT: Still I'd like to thank and congratulate @CharlieLaub for going the extra distance and making a test rather than just talking like me!
Probably audibility is related end result time-domain waveshape, along with the base/fundamental frequency. Purifi has some waveforms on their website that can be listened to. Each one consists of the same 4 frequencies. Both look exactly the same on an FFT. Merely changing the phase of two of the four frequencies changes the result between amplitude modulated or else frequency modulated. https://purifi-audio.com/blog/tech-notes-1/doppler-distortion-vs-imd-7
Beats are envelope amplitude modulations caused by summing multiple frequencies. IOW, beats are not new frequencies, since summing of frequencies is a linear operation.
EDIT: Can you hear the frequency modulation version too? Some people have more trouble with that one.
EDIT: Can you hear the frequency modulation version too? Some people have more trouble with that one.
Last edited:
As an example of what group delay can do to the time domain waveform, here is a comparison between the 100Hz signal with the maximum adulteration I used, and the original signal.
The frequency response (magnitude of all frequency components) is exactly the same for each waveform. The difference between these two waveforms is completely due differences in WHEN each component exits the filter WRT the other components.
These are not all that easy to distinguish from each other by listening. So far I am not getting any statistically meaningful recognition, but the number of returns is only around 20 or so for any of the tests.
The frequency response (magnitude of all frequency components) is exactly the same for each waveform. The difference between these two waveforms is completely due differences in WHEN each component exits the filter WRT the other components.
These are not all that easy to distinguish from each other by listening. So far I am not getting any statistically meaningful recognition, but the number of returns is only around 20 or so for any of the tests.
There still isn't instant switching. There is confusing overlap.
When that's fixed I would be willing to give it a try.
When that's fixed I would be willing to give it a try.
If I implement instant switching in the middle of the test, that probably invalidates it. So I have to wait until the next "round" to do it.
You might also try sweeping the pitch of the test signal as it passes through the crossover frequency. In music this motion would be a glissando. When a "gliss" passes through a phase shifty region, the phasiness may tend to be more obvious.
For me, the hardest part was thinking I had heard a difference in a passage, but due to my extremely short auditory memory (as others have said), I wanted to be able to instantly switch between reference and the option to confidently confirm or deny any difference.
But that would make it synthetic. If played 2 pieces of music back to back, if you couldn't tell the difference, then when it comes to listening to music, does it matter?
I found this a worthwhile test and was as honest as I could be about differences (or not) heard.
But that would make it synthetic. If played 2 pieces of music back to back, if you couldn't tell the difference, then when it comes to listening to music, does it matter?
I found this a worthwhile test and was as honest as I could be about differences (or not) heard.
Took the test. Had to leave most of the music parts choices open.
Thank you for making it.
Was the crest factor in the different tests very different? Had to adust the volume quite a bit.
Thank you for making it.
Was the crest factor in the different tests very different? Had to adust the volume quite a bit.
Thanks for the two paper links. I'm familiar with the Aalto University paper, but hadn't seen Scott's. I can resonate with them, particularly Scott's.
I think it's a good plan to head to a 3-way next. Like wesayso, I think speakers are the better test.
I find the tactile feel, a feeling of rhythm that's imparted by the body participating with the ears is the test I go by. Only speakers can give that ime.
90% of my xover phase experiments have ended up being with the sub-to-main xover in the 100Hz vicinity, which I've found effects tactile feel the most.
IME, its often possible to notice the difference in a once-through play, but usually in a test situation you have to listen and decide over and over again, and try really hard never to get it wrong. Once you do that test/practice enough times, then you very well may start to notice something that goes by very quickly in a piece of music, and that something might a sound your brain totally ignored before. Regarding practice, it may help a lot to get immediate feedback each time you give a right or wrong answer. It can help to give positive reinforcement for correct recognition.
Last edited: