I don't remember which is which, I'll have to go back and check.One file has phase shift, two are the original.
No spoilers, I haven't had a chance to play them yet...
Two reasons I suspect.
1) Many people have long ago made up their minds whether phase shift or group delay are audible or not, probably without any proper, double blind controlled testing, (or any testing at all) and whether conciously or subconciously they are not willing to perform the experiment and thus disprove their personally held beliefs, whichever side of the debate they may fall on. (There is little motivation to go out of your way to disprove your own beliefs)
Or if they have performed the test and it doesn't give their expected result they're not willing to post "looks like I was wrong" in a public forum, especially if they have been publically vocal on their belief on the forum in the past.
I suspect many people with an engineering perspective or background find it particularly hard to reconcile the fact that phase shift (which clearly makes a waveform look different) may in fact be inaudible in most circumstances, and continue to take it as an matter of faith that phase shift is bad without empirically proving it. Too bitter a pill to swallow perhaps...
2) Nobody wants to be seen as having "cloth ears" in a public forum. If anyone admits they can't hear a difference and then other people come along and say that they can hear a difference, now you're seen as the person that obviously has substandard hearing or speakers...(eg its a matter of pride)
You've already got a good clue that gives you a 1 in 3 chance.
I'm not actually too concerned about which is the original and which is modified, only whether the difference is audible or not. The original recording would have had group delay in it too, on each different component before mixing. For example most vocal mic's roll off below 100Hz (or the same is done during mixing) and that would obviously introduce group delay. No recording is free of group delay.
So really we're just comparing two recordings with different amounts of group delay, not group delay versus no group delay.
I did have a quick listen to the recordings on in-ear earphones just manually switching between them and I couldn't notice a difference. I will give it a try on speakers as well though later.
One thing I've noticed that has been indiscriminately mixed together in this thread is group delay versus excess group delay. The part of the group delay that comes from the minimum phase amplitude response is inseperable from the amplitude response - change one and the other changes.
So all the talk about the different group delay of a sealed box versus a bass reflex etc is irrelevant to the discussion of crossovers and vica versa, since the amplitude response is also different, (both sealed and bass reflex boxes are still minimum phase) so you cannot prove that its the group delay that you're hearing the difference of rather than the amplitude response. We're very sensitive to amplitude response variations, it's nowhere near proven that we're sensitive to excess phase/group delay variations to nearly the same degree.
The only way to tell if group delay by itself is audible or not is to look only at excess group delay - eg group delay which is due to excess phase where no change in amplitude response is present - such as your test, and such is the case in a properly summed crossover with a flat amplitude response.
In this case its a lot harder to prove that we can hear a difference, certainly if the excess phase change is smooth and gradual. These results can't be extrapolated to the minimum phase group delay of different bass alignments though.
Personally, if it turns out that smooth excess phase eg excess group delay without large peaks is not audible or only under exceptional circumstances then that's a huge relief, as the only ways of avoiding such excess phase in the summed response is simple 1st order crossovers with all their ills or complex dsp linear phase systems which inevitably introduce pre-ringing and potentially worse off axis ringing.
It's one less thing to worry about that then allows much greater optimisation of other parameters.
Last edited:
or may it's a trap !
3 files are identicals...
check with real time analyser,a nice dip would appear in the crossover region.(but no more delay in the lows )
If the filter that was applied was a perfect all pass filter there would be no dip in the amplitude response in the crossover region...that's the whole point of an all pass filter. Only the phase would be different, so you would need to compare the phase of the two recordings to find which had excess phase. (Something that could be done with the right software)
The questions "is group delay audible" and "what's the audibility threshold at low frequencies" are quite different from the question "is it an audible problem in a real room with real speakers"?
Like some have pointed out, in a typical listening room the group delay from speakers may not be a problem because the room dominates the sound field and the perception of low frequency sound.
Of course, rooms and speakers vary greatly. Nevertheless it might be a good idea to test the audibility or threshold with decent headphones/earphones, and then do a separate test to see if there still is an audible difference with speakers.
Like some have pointed out, in a typical listening room the group delay from speakers may not be a problem because the room dominates the sound field and the perception of low frequency sound.
Of course, rooms and speakers vary greatly. Nevertheless it might be a good idea to test the audibility or threshold with decent headphones/earphones, and then do a separate test to see if there still is an audible difference with speakers.
Yes, ditto the above. 
The room typically dominates. But what if the sound at least comes out of the speakers with little or no phase shift? Will that be an improvement?
Yes, Simon, I believe you're right. No one likes to look the fool. Since I'm already quite foolish, I tend to jump into these sort of tests with both muddy feet.
Also right about the dip. As stated before, I spread the crossover frequencies by 25% to reduce the 3dB bump that Butterworth gives at the x-over frequency. There is a tiny bit of ripple, but I can't imagine it would be audible.
Interestingly, if I mix the 2 files together, you can really hear it - like an echo effect. But alone? I don't notice it. I'm working now to correct phase only in my speakers. The phase now isn't awful, but it's far from perfect. With the convolving tools I should be able to make it near perfect. What will that sound like? I'm not sure. I would hope that flat phase will help imaging, but that remains to be heard.
I may post some new files with that have been high passed like a speaker box, but with flattened phase.
The room typically dominates. But what if the sound at least comes out of the speakers with little or no phase shift? Will that be an improvement?Yes, Simon, I believe you're right. No one likes to look the fool. Since I'm already quite foolish, I tend to jump into these sort of tests with both muddy feet.
Also right about the dip. As stated before, I spread the crossover frequencies by 25% to reduce the 3dB bump that Butterworth gives at the x-over frequency. There is a tiny bit of ripple, but I can't imagine it would be audible.
Interestingly, if I mix the 2 files together, you can really hear it - like an echo effect. But alone? I don't notice it. I'm working now to correct phase only in my speakers. The phase now isn't awful, but it's far from perfect. With the convolving tools I should be able to make it near perfect. What will that sound like? I'm not sure. I would hope that flat phase will help imaging, but that remains to be heard.
I may post some new files with that have been high passed like a speaker box, but with flattened phase.
i've understood that Pano only shifts the lows.
i'm using rePhase as inverse all-pass/all pass in a multi convolution engine.
it allows to manage time independently of frequency response.
rephase
Last edited:
I've made a fool of myself enough times now that I'm not worried about it either.
If you go back through the last 2 years of my posts you'll see that I've changed position on a number of my "beliefs" in that time largely due to the discussions that have made me rethink my position. There's nothing like having to defend your position in public to make you question and rethink it - if you're open minded enough to accept that you could be wrong...I'm curious why you didn't just use a summed L/R crossover of the same order ? Perfectly flat summed amplitude response with no messing around with spreading the crossover points... and more representative of actual crossovers in use. (Who nowadays would use an even order butterworth instead of a L/R in a speaker ?)
So, a 12dB/oct butterworth vs a 24db/oct with post filter phase linearisation applied to both ? Now that would actually be quite interesting...
I would try at least a couple of different cutoff frequencies. Maybe 40Hz and 60Hz ? One to represent a larger speaker, the other for a bookshelf design.
Remembering of course that the cutoff of our own speakers / headphones we do the comparison on are going to impress their own rolloff on top of this...(hence not wanting to go too low with the first one causing the listening speakers/headphones to dominate the comparison)
You'd want a sample song that actually had some low bass in it too. Maybe it was my earphones but I didn't hear much in the way of low bass in your previous sample...
Last edited:
Which is what Andrew was getting at. Even good transducers and circuits aren't perfect in the bass. That's going to color what you hear in the test files.
I can try some other stuff, can anyone suggest a music sample to alter? The Steely Dan had pretty good energy below 100Hz, but not all the way down to 20. Not many do.
I can try some other stuff, can anyone suggest a music sample to alter? The Steely Dan had pretty good energy below 100Hz, but not all the way down to 20. Not many do.
Happy New Year from Portland Oregon USA!
Since you can't separate group delay from time shift (phase shift)(?), I'm not sure how valid a GD variation test can be. Phase shift or time shift will have consequences of their own. Left to right image location is very dependent on timing comparisons left to right and vice versa. Absolute (both channels in sync) time shift relative to the upper frequencies may or may not have perceptual consequences unless the delay is large (many periods).
Since you can't separate group delay from time shift (phase shift)(?), I'm not sure how valid a GD variation test can be. Phase shift or time shift will have consequences of their own. Left to right image location is very dependent on timing comparisons left to right and vice versa. Absolute (both channels in sync) time shift relative to the upper frequencies may or may not have perceptual consequences unless the delay is large (many periods).
Group delay is the first derivative of phase. In other words the magnitude of group delay at a given frequency is proportional to the slope of the phase at that frequency. Nothing more, nothing less.
How can it not be valid ? Add variation in the excess group delay (for example an all pass filter) and see if the effect is audible. That's the crux of the matter, is it audible or not, and what are the thresholds if it is...
Indeed, however I think its implicit throughout all the discussion in this thread that any variations in group delay/phase etc are applied identically for left and right channels.
Again, that's the crux of the issue. I would also like to draw a distinction between peaks in the excess group delay that typically occur at crossover frequencies, (due to the all pass nature of the summed crossover sections) and a shelf in the excess group delay caused by acoustic centre misalignment of drivers.
A group delay peak causes a slight delay in the propogation of the signal over a relatively narrow frequency range (more or less the crossover overlap region) while a shelf in the group delay (such as a tweeter too far forward) will cause a delay potentially over many octaves. (All the octaves below the crossover frequency in a 2 way with tweeter to far forward for example)
That may have a lower threshold of detection particularly in terms of effects on imaging. (Just speculating on this point)
Last edited:
electronica for sure. Aphex twin? Orbital? Im fairly sure ive got something with with sub bass in it. I havent tried to use the testfiles, ive decided that i need a better soundcard since comparing an album in wav, and the same on CD.
Try Yello's Touch album. Not sure it's 20Hz but is pretty low.
Other goodies - Lionell Rogg Toccata and Fuge - 1st track. Track 10 of the Drottningholm Baroque Ensemble (using St. Peters Stockholm, Church organ as bass - Before I got the tapped horns you don't really hear the deepest bass note.
These may be on Spotify.
I have had my Tapped horns (20-100Hz) both time aligned and not, and can't hear a difference to be honest. Using Holmimpulse square wave first positive, I measure that the tapped horns are ~5m behind the bass horn when not aligned.
DCX aligned the tapped horns perfectly with the bass horn- which is 100 - 330hz.
The bass horn and up are physically time aligned using Holmimpulse.
In my opinion time alignment is more critical the higher the freq - Necessary bass to mid horns. Absolutely essential on mid / upper horn and tweeter.
Other goodies - Lionell Rogg Toccata and Fuge - 1st track. Track 10 of the Drottningholm Baroque Ensemble (using St. Peters Stockholm, Church organ as bass - Before I got the tapped horns you don't really hear the deepest bass note.
These may be on Spotify.
I have had my Tapped horns (20-100Hz) both time aligned and not, and can't hear a difference to be honest. Using Holmimpulse square wave first positive, I measure that the tapped horns are ~5m behind the bass horn when not aligned.
DCX aligned the tapped horns perfectly with the bass horn- which is 100 - 330hz.
The bass horn and up are physically time aligned using Holmimpulse.
In my opinion time alignment is more critical the higher the freq - Necessary bass to mid horns. Absolutely essential on mid / upper horn and tweeter.
Last edited:
Back when I worked as an Engineering Assistant in baseband video engineering at Tektronix, some of the engineers there (super sharp guys) designed and built a doo-dad that would cause a significant anomoly in GD over a very limited frequency range, and there was a knob that would position it in frequency. They claimed that they could only hear it's effect in the upper midrange. I never got a chance to try it myself. I don't know anything else about the experiment.
My intended point is that I don't know how you can introduce a significant GD anomaly without simultaneously creating a delay effect. How do you separate the two? How do you say the effect is from GD and not TD or phase shift?
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.