GROUP DELAY AUDIBILITY TESTING - please take part!

[charlielaub]

MY GROUP DELAY AUDIBILITY TESTING IS NOW LIVE
Please use the link in the description, below.

I have launched a suite of online tests where you can see if you can distinguish between various levels of group delay that have been added to several different signals (both synthetic and music types).

These tests are quite difficult, even with the synthetic signals. I have added group delay in levels from inaudible to audible, based on the published literature.

PLEASE ENCOURAGE OTHERS TO ALSO TAKE THESE TESTS.
WHY?
ANSWER: The audibility level for group delay in loudspeakers is an important parameter that puts strict limitations on crossover filter design. Once I have a sufficient number of respondents I will analyze and share the data and full details on the adulterations used to generate the test signals. This information will hopefully be useful to the loudspeaker design community at large.

GROUP DELAY TESTING MAIN PAGE:
http://audio.claub.net/GD_testing/GD_Testing_Main_Page.htm

 

[DWPress]

I saw your other post describing your elliptical filters so I'll look forward to seeing your results.

 

[holdingpants01]

cool test! but it would've been better if there was only one sample playing at one time, now they overlap and play the whole length every time

 

[charlielaub]

cool test! but it would've been better if there was only one sample playing at one time, now they overlap and play the whole length every time

This is the first time I have created a series of tests like this, so I decided to keep the interface simple and easy to code/create. Please put up with this quirk of the testing for now. I can try to improve it, with instant switching, in the future. I have some other tests planned already, if I get sufficient responses this time around.

 

[charlielaub]

BUMP!

I need many more test takers, otherwise it is very difficult to make any statistical observations.

 

[Sylvanus]

What is the frequency-dependent group delay in these tests?
Is it something equivalent to maybe an LR4 crossover in the mid range?
Depending upon the frequency and all-pass characteristic chosen, you could easily adjust the result to be audible or non-audible.

 

[charlielaub]

What is the frequency-dependent group delay in these tests?
Is it something equivalent to maybe an LR4 crossover in the mid range?
Depending upon the frequency and all-pass characteristic chosen, you could easily adjust the result to be audible or non-audible.

While I do not want to reveal too much, I can say that the GD aldulteration that is the smallest of the bunch is equivalent to what you get from an LR4. It should be inaudible on all test signals. The GD goes up from there, and I can personally hear the two with the highest added GD (at least with the synthetic type test signals).

All of the adulterations in this round of testing are with Fc = 1k Hz. This causes any peaking of the GD response to occur around that frequency where the ear is relatively sensitive. In the future I have some plans to test other bands, or multiple allpass type GD effects like you would get with a 3-way loudspeaker (which has two crossover points).

 

[Sylvanus]

I'm not sure what you're trying to suss out here that isn't already known.
These sort of blind A/B tests with included group-delay distortion have already been performed many times.
That said, in some of those tests a variable that should have been accounted for was not. Which rendered those tests invalid.
But, of the ones properly executed with realistic parameters, it's essentially impossible to hear any differences.

In recent times, Linkwitz in his JAES paper in 1976 explored the audibility of delay distortion pretty well.
Other references all the way back to Helmholtz.

 

[charlielaub]

I'm not sure what you're trying to suss out here that isn't already known.
These sort of blind A/B tests with included group-delay distortion have already been performed many times.
That said, in some of those tests a variable that should have been accounted for was not. Which rendered those tests invalid.
But, of the ones properly executed with realistic parameters, it's essentially impossible to hear any differences.

In recent times, Linkwitz in his JAES paper in 1976 explored the audibility of delay distortion pretty well.
Other references all the way back to Helmholtz.

Sure, this is not a new topic and there is plenty of literature already published. What I have read does not always represent the group delay profile from a loudspeaker crossover, and definitely not from the crossovers I am using for these tests since I just developed them and published an article about it. For these the group delay is a bit higher than e.g. LR4 but is below the published audibility threshold. As soon as my publication came out there were immediate questions like "what does the time domain ringing look like". How do you answer that? Of course there is ringing! So does every crossover filter except 1st order, which is not useful as a crossover IMO. But it should be inaudible. How do I convince people that is indeed the case? I thought that gathering some data on the audibility of these exact crossovers would be useful and would help me to answer these questions about the time domain response when other display some concern about it...

 

[OCA]

I tried all the tests. Thanks for the effort. Looking forward to hear the results.

 

[charlielaub]

The tests that have been posted so far require the subject to listen to a few seconds of audio, and then compare that against another track of the same length. This requires some audio memory and since the effects are small and not easily identified, this makes the test more difficult. TBH so far the recognition of any modification, even gross ones, is not statistically significant.

To make the comparison easier it was suggested that I try to implement "instantaneous" switching between the unmodified audio and the modified audio. Using my very basic HTML and Javascript skills I coded up this feature. You can preview it here:

Toggle sound clips

Each line corresponds to one of the modifications. They are for the Pink Impulse and are presented in the same order as the earlier test form. I have not added any voting to the page yet, since I will need to build my own PHP and MySQL code for it. But you should get the general idea.

To compare with instantaneous switching, click on the Start Playback button of any line. The reference audio will start. Which you are listening to is indicated immediately above the buttons. To switch playback between the original and the modified audio tracks, click TOGGLE. The tracks loop indefinitely, so you must click Stop Playback before moving on to another line and a different modification.

This should make it easier to hear the slight differences due to group delay.