I took the test with headphones first. On synthetic samples, it was pretty easy to spot different sounding ones. However, with music samples I’m lost. Any overall difference is on verge of imagination and I don’t know what to look for or how group delay effect sounds.
Using loudspeakers gave the same result. As we are, with rare exceptions, untrained listeners I think that many of us doesn’t know what to look for and subtle differences on music samples are lost.
Using loudspeakers gave the same result. As we are, with rare exceptions, untrained listeners I think that many of us doesn’t know what to look for and subtle differences on music samples are lost.
The literature says this about group delay audibility:
1. There is a threshold (it varies with frequency) for audibility. Below the threshold you cannot tell a difference even if (for a regular signal) you can see that in the time domain that adding the group delay changes the waveform.
2. The audibility threshold is lowest for synthetic signals with sharp transients, and when listening via headphone.
3. Using music signals instead of synthetic ones moves the threshold higher
4. Using loudspeakers instead of headphones also causes an increase in the threshold for perceptible changes.
So, your observations seem to follow all of these trends. As I noted on the main page:
Unfortunately I do not believe that there is a good way to characterize what excessive group delay "sounds like". It would be unrealistic to add so much group delay as to make it very obvious as you might get when adding a lot of distortion to a signal or using a very low bit rate, etc. That amount of GD would never be encountered with a crossover.
1. There is a threshold (it varies with frequency) for audibility. Below the threshold you cannot tell a difference even if (for a regular signal) you can see that in the time domain that adding the group delay changes the waveform.
2. The audibility threshold is lowest for synthetic signals with sharp transients, and when listening via headphone.
3. Using music signals instead of synthetic ones moves the threshold higher
4. Using loudspeakers instead of headphones also causes an increase in the threshold for perceptible changes.
So, your observations seem to follow all of these trends. As I noted on the main page:
But I need to take the same alteration and apply it both to synthetic and music signals to be able to show via the test results these points. Also, I wanted to see if people could detect a lower amount of group delay that I am able to myself, and these may be mostly inaudible. Some of these definitely SHOULD sound exactly the same. I can explain in more detail when I describe the results after I close the testing. For now I am trying to get as many returns as possible in order to get statistically meaningful data.
Unfortunately I do not believe that there is a good way to characterize what excessive group delay "sounds like". It would be unrealistic to add so much group delay as to make it very obvious as you might get when adding a lot of distortion to a signal or using a very low bit rate, etc. That amount of GD would never be encountered with a crossover.
If its okay to say so at this point, just as a general reminder, published auditory thresholds are estimates of a value for the "average ear." They are not hard limits. Thus, we might expect that 50% of the population can't hear below the published threshold, and the other 50% of the population can hear below the threshold. Also, if we assume a bell-curve shaped distribution, we might still want to see if the data better fits a thick-tailed model. https://www.edge.org/response-detail/11715
On the subject of example signals with music, I would suggest to try wide-ish bandwidth LF transient sounds. Deep bass drums with a good beater attack sound, then a nicely audible decay of head resonance modes without further masking by other instrument sounds, and or sound loss by premature gating. Even a higher pitched timbale with a sharp stick attack and the ringing decay might be worth trying. Such sounds can have a sort of time-smeared effect if the percussive transients are stretched out by phase shifts of some frequencies. Crossover frequencies could be mid-band, or even lower between woofer and sub. Also, some systems may reproduce the effects better than others. Depends.
Last edited:
Thanks for sharing in humility…….complex music is nothing like test tones, tone bursts or sweeps. Add in that aural memory for adults is simply terrible past the memory being filled with important sounds required for survival.
Excessive group delay in a crossover should slow the arrival time of low frequency transients relative to the high frequency, harmonics arriving before the fundamental.
Obviously (to me..) with a musical source like castanets Test #4, there is hardly any low frequency, so I wouldn't expect to hear any difference.
I couldn't hear a difference in any of the signals as presented using Sony MD7506 headphones, which have almost flat frequency and phase response up to past 7kHz.
The salsa band Test #6, the only one I'd expect to have a fighting chance of hearing a difference didn't have enough low frequency extension in the kick or bass for me to hear any arrival time difference.
I tried boosting bottom and cutting highs on a second pass through, still couldn't hear a difference.
That said, I have never been able to hear the difference in group delay added from typical 4th order 3-4 way IIR crossovers even when my hearing was still decent, so no surprise I couldn't now that my hearing has deteriorated.
Art
Group delay is common in the carrier section of telecommunications. You may have experienced some times when the signal has a distinct echo or bathroom sound. It can be mild or very severe to the point where you cannot follow the conversation. You cut the link and dial again hoping that you can re-establish the connection without this echo. If you hear anything like this in your audio system, throw it away, the designer of the equipment was a plumber or mechanic.
Thanks for your post with some good points raised. You are correct about (in general) the LF being delayed WRT the frequencies above the crossover point, and I did not take this into account when selecting music tracks. I was hoping to find some stuff, not electronic music but with acoustic ones, that also had some 'transients". I put this all together in only a couple of days so I did not have time to exhaustively search for the "right" music. If you have anything in mind that you think would be a good signal for these tests I would love to use it.
Also, there could be other ways to perform these test, such as the following:
I generate N various levels of group delay representative of the sort of response vs frequency that loudspeaker crossovers create. The N levels should span from below to far above the audible threshold reported in the literature.
The N levels are presented in order of increasing GD to the listener.
The listener is able to instantaneously switch between the reference (un-adulterated) and the signal with group delay added, using buttons placed next to each other on the page.
The listener indicates which level is just audible and that info is returned by the form to me.
I present the test with one synthetic and two music signals only, to reduce the total amount of listening time for a user.
The exact group delay response in each signal is not revealed until after the test, so the user will not know how their own choice corresponds to published levels.
Maybe this sort of test will have the following improvements:
1. instant switching between playing tracks
2. built-in training via the highest group delay level presented
3. a reduced chance of nonsense responses (I am getting some with the current testing)
Thoughts?
I generate N various levels of group delay representative of the sort of response vs frequency that loudspeaker crossovers create. The N levels should span from below to far above the audible threshold reported in the literature.
The N levels are presented in order of increasing GD to the listener.
The listener is able to instantaneously switch between the reference (un-adulterated) and the signal with group delay added, using buttons placed next to each other on the page.
The listener indicates which level is just audible and that info is returned by the form to me.
I present the test with one synthetic and two music signals only, to reduce the total amount of listening time for a user.
The exact group delay response in each signal is not revealed until after the test, so the user will not know how their own choice corresponds to published levels.
Maybe this sort of test will have the following improvements:
1. instant switching between playing tracks
2. built-in training via the highest group delay level presented
3. a reduced chance of nonsense responses (I am getting some with the current testing)
Thoughts?
The more you overthink it, the more you might introduce the same into the results.
It's not about not trusting the participants either. It's well known that there are so many factors beyond our control that testing needs to remain blind.
Like this:
Code:
PLAY LEVEL 1 / PLAY Reference
PLAY LEVEL 2 / PLAY Reference
PLAY LEVEL 3 / PLAY Reference
...
PLAY LEVEL N / PLAY Reference
INSTRUCTIONS : Click the (a radio button, use your imagination) next to the level that you perceive to be just audible. Use level N to hear what gross levels of group delay sound like.
Decades ago, when IIR crossover processor units like the DCX became available, people were eager to try very high order XOs, like 8th-order and (much) higher. And many found out that doesn't sound too well, especially for low frequency XO points.
The excess phase of, say, a 10th order LR at 100Hz should really be audible to most anyone, with a kick drum or plucked upright bass, with the "umph" lagging way behind the "click".
Lower group delays are often audible in the way they can change timbre rather that perceived speed, for similar reasons why flipped polarity can be audible. Also, soundstage perception maybe altered, again for similar reasons why flipped polarity may alter soundstage.
I've sorted out all that stuff years ago and please forgive me for not engaging anymore as this topic is fully settled for me but maybe these pointers might help some.
Training is key, so it is a good thing to start with exaggerated examples to get a feel for it.
Also important, people using regular types -- with significant excess phase -- of multi-way speakers have a penalty from the start, so it is best to use headphones.
And finally, keep listening levels low to very low.
The excess phase of, say, a 10th order LR at 100Hz should really be audible to most anyone, with a kick drum or plucked upright bass, with the "umph" lagging way behind the "click".
Lower group delays are often audible in the way they can change timbre rather that perceived speed, for similar reasons why flipped polarity can be audible. Also, soundstage perception maybe altered, again for similar reasons why flipped polarity may alter soundstage.
I've sorted out all that stuff years ago and please forgive me for not engaging anymore as this topic is fully settled for me but maybe these pointers might help some.
Training is key, so it is a good thing to start with exaggerated examples to get a feel for it.
Also important, people using regular types -- with significant excess phase -- of multi-way speakers have a penalty from the start, so it is best to use headphones.
And finally, keep listening levels low to very low.
Good points all -but I suspect it will have some issues with the converse, as suggested above:
Large amounts of excess are already known to be audible, but using that as 'training' has some issues, since if you provide exaggeration then shift to something else, there is a pre-bias introduced to go looking for it (and then finding it, whether it's actually there or not). Like I often say: I trust my ears, to a point. I don't trust my brain not to get fooled, and it's very easy to kid yourself. Headphones, like with phase linearity, are usually going to be the most revealing of differences like this, but based on Toole etc., I suspect this isn't likely to translate over too well to most practical loudspeaker based systems with their own characteristics, that of the rest of the system & the room to contend with. Not saying it isn't / can't be audible either in some conditions, but the headphones probably can't be used for generalising beyond that [relatively 😉 ] closed system.
Large amounts of excess are already known to be audible, but using that as 'training' has some issues, since if you provide exaggeration then shift to something else, there is a pre-bias introduced to go looking for it (and then finding it, whether it's actually there or not). Like I often say: I trust my ears, to a point. I don't trust my brain not to get fooled, and it's very easy to kid yourself. Headphones, like with phase linearity, are usually going to be the most revealing of differences like this, but based on Toole etc., I suspect this isn't likely to translate over too well to most practical loudspeaker based systems with their own characteristics, that of the rest of the system & the room to contend with. Not saying it isn't / can't be audible either in some conditions, but the headphones probably can't be used for generalising beyond that [relatively 😉 ] closed system.
I bring up training only because some users have asked for it, saying "it would be helpful to know what we should be listening for".
Also, when people listen to sample after sample and hear nothing they might get bored or give up. Even if this is the expected case, it can be frustrating for the test taker...
Also, when people listen to sample after sample and hear nothing they might get bored or give up. Even if this is the expected case, it can be frustrating for the test taker...
If one has to think about what proper sound reproduction sounds like, then they may not have heard it.
Have you ever been bungee jumping? There are those at the top who begin asking intelligent questions about it. The supervisors quickly 'encourage' them to take the leap, before they begin jamming up the stairwells going back down 😉
Have you ever been bungee jumping? There are those at the top who begin asking intelligent questions about it. The supervisors quickly 'encourage' them to take the leap, before they begin jamming up the stairwells going back down 😉
Here I disagree. Some subtle differences are not obvious.
I saw a clip of ABBA in studio yesterday. One of the girls asked: Is it a moog? No its a synclavier was the answer.
I saw a clip of ABBA in studio yesterday. One of the girls asked: Is it a moog? No its a synclavier was the answer.
Thats not group delay, thats when you didn't get an echo canceller device switch in to the call chain... probably because there was no free at call setup or the one you used broke down during call...
GD beings to the "phase" territory... i.e. ms but echo is in the seconds territory...
//
I don't think headphones can ever be a good substitute for how we listen in real life, thus for a subject like this. We listen with more than our ears alone under real life conditions. We feel as much as we hear and it all works together. I like how David Griesinger talks at length about how we perceive things in hist lectures about proximity. He uses a single full range speaker to demonstrate some of those features for a reason. How we take note of things that may put us in danger. Simple first reflections can queue us if that danger is close (to make us fight or flight) or not. Listening to a stereo setup messes up a lot of those queues. So it won't ever be easy to demonstrate any of it but we may have more of a chance when sitting close to a single speaker, avoiding all early room reflections etc.
I've experimented a lot with the audibility of phase myself (using speakers). One could test a signal where the low frequencies arrive earlier than the higher frequencies, I found that much easier to detect even with music. We are far more forgiving when listening to the lower notes lagging behind the higher frequencies. Yet, in studies that Griesinger has mentioned it has been said that we are sensitive to phase even at higher frequencies. It is part of our survival mechanism. But stereo reproduction itself is flawed enough to hide just about any of that.
The cross talk itself, hearing the left speaker with the right ear and vice versa, will hide any phase change and if that isn't enough, the room around us will mess it up.
One of the Griesinger video's that briefly mentions audibility of phase at higher frequencies:
There are lots of other useful video's from him as well as many papers on the subject.
I've experimented a lot with the audibility of phase myself (using speakers). One could test a signal where the low frequencies arrive earlier than the higher frequencies, I found that much easier to detect even with music. We are far more forgiving when listening to the lower notes lagging behind the higher frequencies. Yet, in studies that Griesinger has mentioned it has been said that we are sensitive to phase even at higher frequencies. It is part of our survival mechanism. But stereo reproduction itself is flawed enough to hide just about any of that.
The cross talk itself, hearing the left speaker with the right ear and vice versa, will hide any phase change and if that isn't enough, the room around us will mess it up.
One of the Griesinger video's that briefly mentions audibility of phase at higher frequencies:
There are lots of other useful video's from him as well as many papers on the subject.
Last edited:
IMO headphones are probably the best listening device to use. The true intention (of mine) for these tests is to try and figure out what sort of group delay response is audible, when the GD has been created by the crossover filters. Headphones have relatively flat group delay but loudspeakers already include the group delay of their own crossovers so you are really listening to both sources of GD (from the signal and the device reproducing it). The GD of the loudspeaker is a confounding variable, and everyone will use different speakers, each with their own GD response. The variation between headphones is much smaller and GD is less overall.
By using all-pass filters to generate the group delay response that is added in, one can come up with all sorts of GD response "shapes" but not necessarily ones that correspond to the GD from a crossover.
By using all-pass filters to generate the group delay response that is added in, one can come up with all sorts of GD response "shapes" but not necessarily ones that correspond to the GD from a crossover.