Phase Shift Audibility CHALLENGE

[Amphiprion]

Hello,

About a month ago I posted a short project for those interested in the audibility of phase shift induced by a typical 2KHz passive crossover filter of various orders. There were problems with the processing of the initial data files.

I have gone back and redone all the data file processing, and investigated the post-processing results to verify accuracy. The project can be found here:

http://mark.hayenga.com/audio/phase_shift/

The data files include the original, unmodified song snippet labeled as such. The other five song snippets in each zip file are 'unknowns' to the listener and are given a letter designation. You are welcome to do your own blind listening tests to try and see if you can detect any changes.

I'll post the answer key after people feel they have had enough time to listen to and evaluate the samples.

Hope you guys enjoy!

[Mark K]

Now that's really cool. Thanks Mark. We'll see how I fare...

[cjd]

I'm going to have to give this a shot for sure! Not sure when I'll find the time, but it looks interesting.

[Davey]

Thanks Mark.

Dave.

[Deward Hastings]

Oh, geez, how any hours are there in a day . . . :twisted:

Only an hour or two into the . . . exercise . . . and there are at least a couple in each set where no difference is obvious (to me) after several back-and-forths. I take that to mean that LR2 and LR4 are probably "safe" . . .

(listening so far with Grado SR60 (and not very loud), will try with Shure ER2 next . . .)

Thanks (I think) . . . :lol:

[Davey]

Mark,

I don't understand the pole/zero plots.

In all cases there should be a pole/zero pair that are reflections of each other across the imaginary axis, and located on the unit circle, no?

This must be my misunderstanding of the way Matlab is plotting the results. My Matlab skills are minimal so it probably wouldn't help me much, but this portion of your Matlab source code seems to be missing.

I've listened to the Natalie Merchant and Miles Davis tracks with my headphones and (like Deward) there a couple of them that are indistinguishable from the original.

Good fun, thanks.

Cheers,

Dave.

[Notorious_AK]

I couldn't detect any difference. er-4 headphones. Is there something specific I should listen for?

[Davey]

You shouldn't be able to hear any difference.....in theory.

Ohm's acoustic law says...essentially...that our ear/brain is not sensitive to waveform distortion introduced via phase-shift changes.

For example: A 4th-order Linkwitz-Riley crossover at 2000Hz exhibits a non-constant group delay varying between approximately 220uS and zero over the audio range. The overall phase shift varies from zero to 360 degrees (180 at 2000Hz) over the audio band.

The type of exercise constructed by Mark (or a hardware circuit) is the only way for a subjective evaluation of this phenomenon to be evaluated that eliminates all other variables.

The next step...if we can talk Mark into it ...is to repeat the processing with a lower xover frequency of say 200Hz.

Cheers,

Dave.

[Amphiprion]

Hi Dave,

I don't understand the pole/zero plots. In all cases there should be a pole/zero pair that are reflections of each other across the imaginary axis, and located on the unit circle, no?

Well, in the Z-plane, the unit circle is effectively the same as the S-plane imaginary axis. For all-pass filters in the Z-plane, each pole/zero pair 1) occurs at the same angle theta from the Z-plane origin and 2) their distances from the origin are reciprocals of each other. So in polar coordinates, the pole/zero pair will be at (theta, R) and (theta, 1/R) where R is the radial component of the pole. Since the pole has to be inside the unit circle for stability, R is less than 1 for the pole and greater than 1 for the zero.

The code for generating the impulse response plots and pole/zero plots wasn't included in the matlab code because I used a built-in function, the filter visualization tool (fvtool.m) to generate those.

The next step...if we can talk Mark into it ...is to repeat the processing with a lower xover frequency of say 200Hz.

That's actually in the plans, though I was thinking of 300Hz But I was going to let the 2kHz test simmer for a while before I cooked up another batch of tests.

Good fun, thanks.

Happy to do it. I figure everyone donates what they can to the DIY community, whether it be loudspeaker designs or things like this. This is just my little contribution.

[Johnloudb]

If some people get happy feet for the results you could PM them the Key to the different tracks.

I'll take this test shortly. This should be very informative, I think.

[1Michael]

I could hear no differences.

[Amphiprion]

Any other experiences?

[jkrutke]

Any other experiences?

I was able to tell them apart - by cheating with diffmaker. :B It was a strange effect, but I would guess that the difference files with more content were the ones with more phase shift. And it wasn't like you could listen to the results and say "oh, so that's what phase shift sounds like" - it was just a vague midrange content.

I only did the diffmaker thing because I couldn't tell any difference by ear. On the other hand, my headphones aren't very good.

[Amphiprion]

I was wondering how soon it would be until someone would cheat

[Davey]

I cheated too.....but only after listening.

I only listened to the Merchant and Davis tracks with my ER-4's. It's very difficult to reliably identify a difference, but I think I could with two of the tracks on each selection. I assumed those were the 4th and 5th order ones.

Cheers,

Dave.

[---k---]

I tried it with a pair of Shure SE210 headphone. I couldn't tell a difference.

Like Zaph, I then ran Diffmaker on them. I thought maybe if I heard the differences I would be able to go back and listen for it specifically. I was shocked at how much content there was in the Diff file. There really wasn't one particular thing that you could listen for. I then listened again, and Diffmaker makes it really easy to flip between files, and still couldn't tell a difference.

[dmalphur]

I used diffmaker to A-B compare the files and could hear no difference. Like others, I was surprised at the amount of information in the difference file.

-David

[mazurek]

I refuse to download your songs because I am recovering from audiophile psychosis already.

Anyway, here is my experience doing very similar experiments with the freely available (for about 15 minutes) program 'Phase Arbitrator'.

When I set it to correct my LR2-20Hz, LR4-140Hz, LR4-1700Hz system, both myself and my roommate could consistently pick the corrected one as the better one (basically with a verbally controlled toggle between an unknown 1st and 2nd, listener tells toggler to stop on their favorite option, high number of consistent picks for the favorite being indicative of a difference). Music that was helpful was generally percussion, or impact instruments (cymbals and some types of drums). We did the same experiment except for just correcting one or two frequencies at a time, and found I believe that the lower frequencies had a more noticeable effect. Then we went as far as just correcting one of the allpass' at 140Hz to make it equivalent to an LR2. Much more difficult to tell the difference. I think the moral of the story was that it can make a difference if the entire time delay is corrected, but I didn't feel that simply going from LR4->LR2 made all that noticeable a difference.

Later, I did the same experiment with Mazeroth. This one was all or nothing, either the entire crossover was phase corrected, or the correction was bypassed. Same method was used, toggling between an unknown 1st and 2nd and then stating a preference for the current selection (several times in a row with randomly changing 1 or 2). This time, I think we both picked the uncorrected as favorite, though I don't remember exactly. My take away from this session was that my crossover was changed between sessions and for some reason that affected my phase preference.

I have noticed that driver to driver phase (as opposed to system phase) can be responsible for some perceived brightness, when I hear the highs disconnected from lows it can manifest itself as brightness rather than one percussive impact.

[Xander]

I'm going to try this out with my Sennheiser HD-280 Pro headphones (first pair of decent headphones, I love the sound of them). I'll report back when I get a chance

[Wilk]

That is a really cool test Mark.

What I have been wondering though is how those tracks would sound should just a minor deviation from any of the target slopes effect things.

As in introduce a small % error into the slope of either the high or the low pass section in each of the filters. Then how badly does this error present itself in different order slopes.

It would be interesting to see what filter slopes are more forgiving/picky about minor misses from the acoustic target. I suspect higher orders show more issues, but I have no idea how the target miss would effect the actual sound as we hear it.

Knowing target slopes are nearly imppossible to hit accurately with driver responses, and 5% componets, it would useful to know how close you have to be before it really starts to screw things up.

[Johnloudb]

Hi Mark,
Well I did listen to the Natalie Merchant tracks in a couple listening sessions. I thought difference on at least one of the tracks but I'm not positive. I think in longer listening sessions these differences may be audible, or maybe not. I didn't use diffmaker on them though. I'm curious what that sounds like. I'll probably try that when I get a chance.

[ch83575]

Very cool post... thanks for the test Mark. I, like everybody else, have a very hard time noticing any difference. I see a lot of comments about the headphone quality used for the test, but I think that is missing the point. Any headphone of reasonable quality will be better in this test than even the very best dynamic loudspeaker. Loudspeakers have acoustic centers that are inches apart, are meters from the listener and every single frequency is subject to non-linear room interaction. Speakers are a phase coherency nightmare, so I think this test shows one thing for sure: the PHASE SHIFT is not what we should be worried about. I think that it would be easy to misinterpret that though... we should not discount all phase issues. I too would love to hear a test with a crossover that deviates from a textbook slope. But, the thing that I think is much more important that each of those issues is phase matching. I suspect that it is possible to hear when drivers deviate from perfect phase matching even far outside of the passband. I don't know how difficult this would be to simulate in matlab, but I think it would make for another cool challenge. My guess would be that instead of being surprised by how little change there is we would all be surprised by how little deviation it takes to be audible even octaves outside the crossover. Of course its only a guess

[Undefinition]

No veils lifted for me

I tried the files in diffmaker, but am unsure about the "extracted" results--seems to be too much difference, honestly. However, toggling between the original and samples in diffmaker while simply playing back, I couldn't hear a difference.

I think maybe the best way to measure these sort of subtle differences are with clips that are only 1-3 seconds long. I guess it's because my brain thinks in the world of instruments, and it's sort of analogous to trying out different patches on a synthesizer while playing the same note over and over. Or you could think of it like twisting the EQ knobs while the snare drummer repeatedly hits quarter notes until you dial in the sound you like. At least, that makes more sense to me than listening to a whole clip, then going back and trying to remember what something sounded like 30 seconds ago.

...so, um, where's the answer sheet?

[Amphiprion]

I'm still holding off on posting the answers. I'm afraid once I post it, I'll have a slew of people claiming to correctly have identified everything doing it blind

[Johnloudb]

Mark, I mentioned your phase shift challenge on the DIY forum and a member, who had a phase shift box, mentioned that you can easily hear phase shift once you know what to listen for.

So, I was wondering if you could post a set of tracks with the accurate phase shift labeled on them. This would let people learn the differences with practice, if possible. Then they could test themselves with unlabeled test tracks. If they can't identify them after some practice, then they probably aren't audible.

I thought maybe I heard some difference, but not positive. If it's not too much trouble, I think it might interesting to see if this makes any difference.

[Johnloudb]

Ha,Ha, .... Well it turns out that his phase shift box (effects box) he had doesn't actually do what he thought it does. Well, it'd be intertesting to see if knowing which tracks had different amounts of phase shift could help people learn a difference and then they could test themselves with blind tracks to see if they really can.

It may be more trouble than others what to bother with though - kind of time consuming.

[cjd]

I'm hearing incredibly slight differences after listening over and over and over to Natalie Merchant on random repeat. Working through the other tracks right now also. Mostly it's audible to me in the voice - it's higher frequency content where it seems she's actually very slightly overdriving the mic (or there's a compression artifact of some kind somewhere in there).

It's definitely not something that folks preferring the sound of MP3's or with standard muck and distortion artifacts in their system are likely to notice. It's not something I probably care about. Hearing a difference after listening to tracks over and over could entirely be a mental attention difference, so I'm not even convinced I'm hearing anything at all.

Most of the difference is not in the core sound, or in the particularly high frequency sound. I was guessing a 2500Hz rotation center, slightly off perhaps. (I looked at the graphs eventually).

It's not high frequency enough to mess with spatial sense (something I know I'm particularly sensitive to), and not low enough that it catches most core musical frequency output.

How hard was this to set up, Mark? I'd LOVE to hear some with a 400Hz center point.

On the other hand... I've not ever been one to think that slope made as much of a difference as how it is used and how well things integrate as a result.

C

[Amphiprion]

It's very very easy to set up, just tedious to work through all the files and make sure everything is done correctly. I'll work up another set with a lower crossover point, though I was leaning toward 300Hz since most three ways try to make the midrange cover 300-2k or 300-3K.

[cjd]

That works. My frequency range baseline is A440 from the years playing violin.

I suppose this is unfair - I already know that I hear no particular issues listening to solo violin on my 3-ways which use steep filters AND have crossover points such that a single instrument covers the full range across all filters.

I'd also be vaguely interested in listening to the current crossover point with speech rather than pure sound.

All that said - our brains are quite adept at filtering through this kind of phase issue.

C

[evilskillit]

Well I'm not really sure what this is all about but I happened to be sitting here at my computer with my HD600s on, hooked up to a decent headphone amp so I figure I'm in a good position to do some listening tests.

For the most part just listening once to each group, listening critically but not overly so, as if I were just listening for the sake of the music and not to pick it apart, I could not really tell any difference at all between any of the Natalie Merchant stuff. The Miles Davis clip was very annoying with the horn going only in my left ear for about 2 minutes, clips D and E seemed less annoying but maybe I was just getting used to it.

Alice In Chains I was familiar with so I was able to get a bit more out of listening to those files. For the most part they all sounded the same. A/B comparison showed me there might be some small differences especially in the R and T files which sounded more spacious, like the instruments were moving around a bit. However if you gave me any of these files and did not tell me they were modified I would never know the difference.

[Johnloudb]

I think most people are not going to hear a difference right off the bat. Familiarity and experience matter. I'm going to listen more to these see if I can pin point any differences. To really confirm that there is an audible difference, you learn the differences of the tracks over a period of time. Then you test yourself to see if you can accurately identify those track you learned.

Just like school.

Anyone can do this by simply duplicating the tracks (test tracks) and then have someone else rename the track files, and write down the "Key" to the original tracks.

So, if you can pinpoint some audible differences over repeated listening over say a week or two. Then try to identify the test tracks. If you accurately identify the test tracks, then that proves audible difference. If you can't then there is likely no audible difference.

[cjd]

Forgot to mention (as if it matters) - tested with Ultimate Ear Pro5's and fresh foams.

C

[Johnloudb]

I think speaker builders have better trained ears , although the UE Pro5's certainly don't hurt.

The best I can do is "sense" a difference listen to Natalie Merchant tracks. I can't say what the difference is. I'm listening on Sennheiser HD600 headphones and a Headroom Little headphone preamp plugged in a E-MU 0404 low distortion sound card. Though, I think the listener probably matters more than the equipment.

I'm gonna keep listening each day. See if I can really identify anything different between the tracks.

[cjd]

I think trained ears helps. I've been surrounded by music (classical and jazz) since before I was born - asked to play violin at 2, finally found a teacher when i was 3... I doubt that I could pick out the difference consistently and say "this track is this crossover", ever. It's NOT that kind of difference I hear. At least yet. I listened on random repeat till I was unable to take it any longer and never landed on anything more concrete.

[Johnloudb]

I think the way to identify these differences, if possible, is to compare the different tracks to the reference track with no phase shift. You should know which track your listening to and go then go back and compare to reference track occasionally. Once you get where you think you can identify the tracks, after a week or two, then try random repeat and see if you can still identify them. That's the test.

You could do it for a short period each day, learning the difference, and even take notes of the differences you hear if any.

This is a pain in the butt!!! But, it would be really interesting to see if you can do it. It's kind of difficult for me to do these tests because of this ear sensitivity problem I'm dealing with. I can't always listen that long to music, some days.

Charles Hansen did blind testing of relays and switches like this. Here's a blind test Charles Hansen did.

This comes from the DIY Audio post Here

Charles Reply - Here

I not concerned with the difference a switch or relay might make, but I won't say it's not audible after the experiences I've had with my hearing. But, I do agree with his testing strategy.