try the second example as it is more obvious there, listen at a very high volume with your headphones and to a short segment of maybe 1-2 seconds (i used a clean note of the saxophone, or whatever this instrument is called). listen to it while alternating between version a and b for about 5-10 minutes. if you dont get mad while doing this, you will find a difference this way, at least thats how i got it.
I've been doing exactly this for an hour 
I can not say I'm sure there is a difference neither on songs 1 and 2.
One important thing to note: Looking at the time domain waveforms again one can see there is difference in peak amplitudes on some notes (for example on the saxophone) about 1dB or so and it is due to the delay difference of low vs high freqs. Now, if you use high volume at playback the nonlinear distortion will be different and it can be audible. How one can be sure it is the phase difference one is hearing and not something else??
- Elias
I can not say I'm sure there is a difference neither on songs 1 and 2.
One important thing to note: Looking at the time domain waveforms again one can see there is difference in peak amplitudes on some notes (for example on the saxophone) about 1dB or so and it is due to the delay difference of low vs high freqs. Now, if you use high volume at playback the nonlinear distortion will be different and it can be audible. How one can be sure it is the phase difference one is hearing and not something else??
- Elias
Thats true, Elias. Replaygain indicated a difference in average levels of about 0,01db, which could obviously be more with the peaks. So, if the LR24 made peak levels differ additionally to the phase change, the whole test is flawed. On the other hand, since he used a normal filter for this test, we can assume that the filters in our crossovers do the same amplitude changes. So, even if it isnt a valid test for phase errors, at least its a valid test for phase related crossover distortion.
What i have learned from it is, that phase could very well be an overblown concept. Since i can hardly detect the changes a 24th order crossover brings with it, how can i ever detect the changes a normal crossover does?
What i have learned from it is, that phase could very well be an overblown concept. Since i can hardly detect the changes a 24th order crossover brings with it, how can i ever detect the changes a normal crossover does?
Last edited:
I've been doing exactly this for an hour
I can not say I'm sure there is a difference neither on songs 1 and 2.
One important thing to note: Looking at the time domain waveforms again one can see there is difference in peak amplitudes on some notes (for example on the saxophone) about 1dB or so and it is due to the delay difference of low vs high freqs. Now, if you use high volume at playback the nonlinear distortion will be different and it can be audible. How one can be sure it is the phase difference one is hearing and not something else??
- Elias
This is a good point. I may have gone a little too heavy on the order of the allpass causing very small amplitude distortions that could give them away.
I actually did check the examples themselves which I will show you how to do later. I could have gone a little lower on the order and it would have been totally clean. I must admit there are probably a few small amplitude distortions in there.
Still when compared to what you would expect when seeing the measurements I was sort of shocked. The first time I played the distorted file I was afraid to turn the volume up so I started at muted and slowly turned it up.
This is similar to my experience with ABXing. And yeah it does help to have some sort of idea what to look for. Really in this case I am not sure what to look for in transient distortions.
There is an entire signal chain in this test and a few intermediate steps in the chain that may have messed this test up the more I think about it. Basically I am not doubting you or your friend but there is a small possibility that the difference you heard was another distortion that was added - I used dither.
Maybe not flawed at all, any filter with non-linear phase will alter the envelope of a complex waveform (multiple frequencies). No way to get around this, so it is a package deal when altering the phase relations between different frequencies.
Well I am not sure it's the filter so much as the limitation of 44.1 16-bit which makes the test more accessible. But makes keeping the distortions to a minimum from the filter not exactly trivial for me. The signal chain is as follows.
EAC secure rip+test and copy> 44.1-32-bit float>LR24>Highpass Triangular Dither (this is frankly not the best option maybe a little lazy on my part if I am honest) 44.1 16-bit
The other one was left untouched. So in an attempt to keep distortions to a minimum I may have given it away with the dither and also yest there are very small amplitude distortions in the audible range which may have been nonexistant in a higher sample rate (not sure). *edit On test tones though the THD was around -140.
I really didn't go at these ABX style myself. I probably could ABX them I have done it with other things where I would actually in retrospect say for the most part the difference isn't noticeably audible.
EAC secure rip+test and copy> 44.1-32-bit float>LR24>Highpass Triangular Dither (this is frankly not the best option maybe a little lazy on my part if I am honest) 44.1 16-bit
The other one was left untouched. So in an attempt to keep distortions to a minimum I may have given it away with the dither and also yest there are very small amplitude distortions in the audible range which may have been nonexistant in a higher sample rate (not sure). *edit On test tones though the THD was around -140.
I really didn't go at these ABX style myself. I probably could ABX them I have done it with other things where I would actually in retrospect say for the most part the difference isn't noticeably audible.
Last edited:
I did some listening today. I think I hear a difference. One cut seems to have more ambiance and reverb than the other. Can hear it on Lou Reed's voice and also on the sax. More sense of space. Kick drum seems different too.
If I could get the ABX thing working, I could test myself to see....
If I could get the ABX thing working, I could test myself to see....
I would fail that test for sure
I was giving people a fighting shot by there being two things which they could contrast that in theory the only thing wrong is the phase. But yeah this test could be arranged. I was actually going to just label them clearly and only accept ABX or no difference reports. But to me some of the most informative stuff about ABXing is the person who is doing the tests impressions of the differences and if they were huge or very small etc..
I am going to do some thorough real world measurements this time before I commit to an entire signal chain. There are a lot of different ways to go through the filter and I really didn't consider all of them. I may also post some high res examples next round see if that has any different results. Possibly might have to reduce the order a notch or two.
Last edited:
How important is a more or less linear phase response in speakers? I mean, it seems to be practically inaudible. On the other hand, for example, the Danley Sound Labs speakers are marketed with their good phase response. So i would guess i am missing an important piece in the puzzle.
Last edited:
Well I dunno if I would jump to a conclusion yet. This is an isolated example of a specific order. For all we know at this point maybe higher order is somehow less audible? Also consider that what I am doing when I combine the HP and LP is in an entirely different domain than the acoustic non coincidental summing that takes place with a speaker.
Another example for people who are curious. LR50 @ 1.5kHz Examples are at 96kHz 24-bit.
zSHARE - Band of Gypsys S1 - 01 - Who Knows A.flac
zSHARE - Band of Gypsys S1 - 01 - Who Knows B.flac
96kHz 24-bit Logsweep (note I do not think it is that pretty with a real signal)
The recording is from 200gram vinyl. This brings up other variables that we are probably not considering as casual consumers. Recordings of the nature that I have provided are most likely no where near being "phase linear". If it goes to a tape it goes through filters etc... So this is a casual test by nature. I may try to go further with this and attempt to make a somewhat phase linear recording and see if in that case it somehow does make a difference more obvious.
zSHARE - Band of Gypsys S1 - 01 - Who Knows A.flac
zSHARE - Band of Gypsys S1 - 01 - Who Knows B.flac
96kHz 24-bit Logsweep (note I do not think it is that pretty with a real signal)
An externally hosted image should be here but it was not working when we last tested it.
The recording is from 200gram vinyl. This brings up other variables that we are probably not considering as casual consumers. Recordings of the nature that I have provided are most likely no where near being "phase linear". If it goes to a tape it goes through filters etc... So this is a casual test by nature. I may try to go further with this and attempt to make a somewhat phase linear recording and see if in that case it somehow does make a difference more obvious.
Linear phase is fairly important, but you have to look carefully. The only phase that is in the least bit audible is non-minimum phase or group delay - more meaningfull because it is in the time domain. The "linear" part simply means a time delay, which is either extremely audible or not audible at all depending on how you ask the question. Group delay is audible under the right conditions, no one doubts that, but what level of group delay is audible is a complex question - because it depends. Group delay audibity will vary with frequency and most importantly level. Group delay is most audible in the range of 1 kHz - 6 kHz and its audibility gets great with greater SPL. So basically at some level is very important and at others not so.
Interesting stuff. If I were making the test files, I wouldn't use such high orders of crossover - how many speakers have 24th order? How many have 2nd or 4th order? Max, I would try 8th order. Maybe process the files with two XO points, like 250 and 2500Hz, LR4, more like a good speaker would be.
Thanks for your work! It's interesting.
Thanks for your work! It's interesting.
Well not so sure that is the only kind of phase change that is audible. Pretty sure asymmetrical distortions would be much more obvious - if I was to say run the left channel only through an all pass and leave the right alone it should be audible.
I think this test is more about transient reponse. In theory I really should be messing up the TR but somehow these time errors even though they are large seem to have no effect on a set of speakers in a normal room.
Thanks. Well I guess the reason I was trying such a high order is I was having a hard time getting much of anything with small phase changes - going from my normal monitors with phase distortions to using an impulse response to linearize the monitors seemed to have only a very subtle effect if it was actually perceivably there at all. So I thought why not really distortion a signal to give an idea of what to be listening for - and of course this was no help.
But later I could lower it to something more real world and see if that somehow changes the results.
Key,
How, exactly, are you processing these files? This type of exercise is not a trivial one and there are some possible pitfalls.
For example, the peak levels of the processed file will be increased and this could result in clipped samples if the peak levels were near 0dbFS to begin with. One of the things to check would be to verify that the processed files don't have any clipped samples. If they do you'd need to reduce the level of the original file beforehand to prevent this from happening. That reduced level file should become the "original" for comparison as well.
Cheers,
Dave.
How, exactly, are you processing these files? This type of exercise is not a trivial one and there are some possible pitfalls.
For example, the peak levels of the processed file will be increased and this could result in clipped samples if the peak levels were near 0dbFS to begin with. One of the things to check would be to verify that the processed files don't have any clipped samples. If they do you'd need to reduce the level of the original file beforehand to prevent this from happening. That reduced level file should become the "original" for comparison as well.
Cheers,
Dave.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.