Crest factor is a likely clue. Volume envelopes may change significantly with selective phase shift changes.
Yes. Peak value can increase by a factor of 3 for square wave (segments thereof, like a clipped kick or snare drum).
Hi Mark
There is no right or wrong here. The test is meant to find the statistically relevant audibility thresholds of a large-enough group of people. Which could possibly lead to the optimisation of crossovers. So there is no correct or incorrect recognition.
I guess all samples from the multiple-choice (maybe except a hidden reference) are manipulated to a specific degree. So again a question of hearing threshold and not of right or wrong.
Correct
And if it is just the different crest factors that we hear this would also mean a statistically relevant result because it means that we should maybe keep an eye on that when designing crossovers. But this means that no other manipulation of any level should be made to the test samples except the one that naturally takes place due to the phase-shift involved.
Just as a side-note: Some infamous voice processor does intentionally use phase shift in order to achieve a more symmetrical wave-shape of male voices such that FM transmitters can be driven harder without running into overmodulation.
Regards
Charles
Group delay slows the arrival of whatever frequencies are affected by group delay. The Measurement data of group delay literally tells you what frequencies are being delayed and gives tell of time till peak energy.
Group Delay is usually accompanied by ringing is well, for example the peak in group delay at a filter knee will also have decay pushed out further in time, late to start and late to finish.
@CharlieLaub maybe add some test where the group delay is described in detail so we know where to focus our ear.
The natural trend of group delay is always decreasing from low frequencies to high frequencies. There might be some peaks and dips (even negative ones) inbetween but the general trend of "naturally occurring" group delay is always higher at low frequencies and lower at high frequencies.
If it is extreme it can make clicks turn into something sounding like "Peeooouuu" (Piiiuuuu for the German speaking).
According to Martin Colloms it can also alterate the perception of dynamics when it comes to loudness differences between transient and static signals. This might propbably have to do with the change in crest factor that has been observed here.
Regards
Charles
If it is extreme it can make clicks turn into something sounding like "Peeooouuu" (Piiiuuuu for the German speaking).
According to Martin Colloms it can also alterate the perception of dynamics when it comes to loudness differences between transient and static signals. This might propbably have to do with the change in crest factor that has been observed here.
Regards
Charles
Some of the examples sounded really "plastic", very easy to detect. Similar sound I've heard with many sound processing plugins like old noise reduction plugins and compressors, really couldn't use those (as in mixing recorded sound). I do not know what is the state with modern plugins, likely better. Similar sound signature though, really obvious.
Some samples appeared somehow brighter, not as relaxed and natural sounding as the reference. This was little unexpected but I think it was real as not all samples did this. It was bit hard test as the samples had to be played until the end before switching. I think I was not perceiving anything timing related per se, as if something was early or late, but more like change in color kind of thing. I tried with headphones and quite low SPL.
Some samples appeared somehow brighter, not as relaxed and natural sounding as the reference. This was little unexpected but I think it was real as not all samples did this. It was bit hard test as the samples had to be played until the end before switching. I think I was not perceiving anything timing related per se, as if something was early or late, but more like change in color kind of thing. I tried with headphones and quite low SPL.
I am working on implementing the testing with instant switching. I have coded up a basic web page for the Pink Impulse that allows you to toggle between the reference (called "Original") and the same N modified (processed) audio tracks as is used on the Pink Impulse page that I posted a few days ago. Please give it a try and tell me what you think about this type of playback control. Note there is no voting, however, the tracks are presented in the same order as on the Pink Impulse page that allow voting so you can open both and compare and vote via the page with voting.
INSTANT SWITCHING VERSION (live demo, without voting):
http://audio.claub.net/GD_testing/test3/PinkImpulse_InstantSwitching.html
I will keep working on this to incorporate voting and text and other elements. Feel free to give me your honest feedback while I improve things. Thanks!
INSTANT SWITCHING VERSION (live demo, without voting):
http://audio.claub.net/GD_testing/test3/PinkImpulse_InstantSwitching.html
I will keep working on this to incorporate voting and text and other elements. Feel free to give me your honest feedback while I improve things. Thanks!
I have been playing with the code I developed to implement the instant switching (link above in post #70) but it will not work with the form javascript that I borrowed from my web host. I will have to create a simple PHP server to collect the responses, but it doesn't seem too difficult. It's my first time using PHP and setting up such a system, so please be patient when my forms look like a 3rd grader wrote them!
Nice try Charlie!
I find it much more convenient now.
Maybe it will be easier now to detect differences in the music tracks.
regards / George
I find it much more convenient now.
Maybe it will be easier now to detect differences in the music tracks.
regards / George
Hi Charlie
The instant switching has one modified version less than the non-instant switching one. Are you sure the rest of the modified versions are really the same for both ?
Regards
Charles
The instant switching has one modified version less than the non-instant switching one. Are you sure the rest of the modified versions are really the same for both ?
Regards
Charles
@phase_accurate Hi Charles, Yes, the new form has the same number of modified audio tracks (six).
if you look closely at the initial form, there are six Option buttons plus one button for the Reference Audio, so seven tracks in total. To hear the reference you have to click the top button.
On the page with instant switching, the Reference Audio (called "Original") is available on each line as one of the two possible tracks that can be played. There are six rows, one for each option (each different modification of the original audio). So there is no need to have a separate line or button for the reference audio.
if you look closely at the initial form, there are six Option buttons plus one button for the Reference Audio, so seven tracks in total. To hear the reference you have to click the top button.
On the page with instant switching, the Reference Audio (called "Original") is available on each line as one of the two possible tracks that can be played. There are six rows, one for each option (each different modification of the original audio). So there is no need to have a separate line or button for the reference audio.
OK - now I see. Sorry for the commotion. Now I also understand why I never heard a difference with the new test setup ! I was always listening to the reference track !!! 😳
Ah, sorry if that was not clear. When you begin playback on any line in the new form it is playing the reference (Original) audio. Only when you click "TOGGLE" does the playback switch to the modified audio. Use toggle to switch back and forth as many times as you would like between the original and the modified audio tracks while they continue to play.
THE FIRST ANALYSIS OF THE TESTING IS HERE !
I finally got a total of 50 returns for TEST # 1, the square wave.
I analyzed the returns using Pearson's chi-square test, with 1 degree of freedom, at a confidence level of 95%, to obtain the statistical significance for each modification of the original audio. You can read more about this statistic and how to do the analysis at the following links:
https://en.wikipedia.org/wiki/Pearson's_chi-squared_test
https://longform.asmartbear.com/ab-testing-statistics/
The results are shown in the image below:
Only one of the modifications is statistically audible using this test method.
The group delay response vs frequency for that modification looks like this:
Based on the published literature, this should be audible, so the result is not a surprise. In any case it is great to have some data at last!
Thanks for participating. The tests are still open and ongoing. I will post if/when I get enough returns for the other tests, e.g. when they reach (hopefully) 50 returns. Currently all have around 35 returns each.
.
I finally got a total of 50 returns for TEST # 1, the square wave.
I analyzed the returns using Pearson's chi-square test, with 1 degree of freedom, at a confidence level of 95%, to obtain the statistical significance for each modification of the original audio. You can read more about this statistic and how to do the analysis at the following links:
https://en.wikipedia.org/wiki/Pearson's_chi-squared_test
https://longform.asmartbear.com/ab-testing-statistics/
The results are shown in the image below:
Only one of the modifications is statistically audible using this test method.
The group delay response vs frequency for that modification looks like this:
Based on the published literature, this should be audible, so the result is not a surprise. In any case it is great to have some data at last!
Thanks for participating. The tests are still open and ongoing. I will post if/when I get enough returns for the other tests, e.g. when they reach (hopefully) 50 returns. Currently all have around 35 returns each.
.
I have done it with headphones but didn't find the time to test with speakers so far. It makes me wonder how me and my speakers will fare together (their crossover-induced GD distortion is slightly lower at DC than your best example and decaying towards higher frequencies without any peaking).
One thing I could clearly see, or better - hear : The perception seems to be SPL dependant like Earl Geddes wrote. When listening at a very low level all examples sounded the same to me. It is when I turned the volume up differences began to show up.
Regards
Charles
One thing I could clearly see, or better - hear : The perception seems to be SPL dependant like Earl Geddes wrote. When listening at a very low level all examples sounded the same to me. It is when I turned the volume up differences began to show up.
Regards
Charles
Been watching this. 🙂 Question Charlie. Seems odd to me that a 0.9ms variation between the DC level and the peak is statistically audible where as a 3ms difference isn't. But heck, I can hear any difference at all. Then again, I don't listen for nuances anymore.
Also, I'm guessing, but it appears the the 2nd case, 0.45ms DC, 0.55 peak, is what you would get with a 1k Hz LR4 crossover.
Audible to what percentage of the population? 50%? 100%? If its a threshold value then it should be audible to 50% of the population (assuming the test subjects have "normal" hearing).
Anyway, in case its not obvious, the first thing you need to do is establish if your test subjects have "normal" hearing. Otherwise your measurements will be completely uncalibrated relative to published measurements.