Phase Shift Audibility Tests

[Amphiprion]

For your enjoyment:

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

[Johnloudb]

Mark,

Looks great! You forgot 1st order. Well, we all know that first order is phase coherent, with prefect waveform preservation (ideally).

And you made a mistake on 4th order phase plot. You posted the second order one.

Very informative though. Thanks for sharing!

[Amphiprion]

Hi John,

Thanks for the compliments (and catching my error on that plot). I threw the website together real quick tonight, I'll get it fixed next time I head back to my work machine. Right now I'm on the laptop getting ready for some TimeWarp and Mythbusters on Discovery If anyone catches any other goofs please let me know and I'll get them fixed.

I skipped 1st order simply because I don't really see anyone using them these days, and also because I wanted to keep everything Linkwitz-Riley based. I would have gone higher than 10th order, but I had problems getting my Matlab code to work at anything higher than 10th order - the pole/zero plots started coming out all wrong and the frequency response was anything but flat. Something about how the built-in invfreqz() function behaves. But it worked great for up to ten.

[Davey]

Mark,

The labeling might confuse some folks.

A second-order LR crossover has a first-order all-pass phase response. (0-180)
A fourth-order LR crossover has a second-order all-pass phase response. (0-360)

Etc,
Etc,

Cheers,

Dave.

[Amphiprion]

You're right, that would make more sense. I'll change the graph titles and make note of the correlation in the text above. I don't know what I was thinking when I typed those.

[chrisn]

Interesting. Thank you for using Alice in Chains I'll have to listen again later, when the dishwasher is not running.

[crackyflipside]

Okay I tested it with Miles Davis and my desktop system. Insignia speakers, behringer a500 amp, rythmik 12" sub.

I got all the files onto the playlist, then loaded the same files 3 times so there were now 18 total songs on the playlist. Clicked the shuffle button a bunch of times. Clicked play and turned off the computer monitor. I did see the first song which was the unmodified, the rest were unknown and in random order. Ranks were giving by or and any additional comments I wrote down during testing will be summarized next to smilies in brackets this was the tally:

Unmodified
Test #1 [control]
Test #11 [I wrote down unmodified]
Test #17 [sounded 'alright']

1st order
Test #4
Test #8 [bad treble]
Test #16 [bad treble]

2nd order
Test #2 [sounded open]
Test #13 [less clear, maybe same as test #3]
Test #15 [reminds me of test #9]

3rd order
Test #3 [less clear, small soundstage?]
Test #5 [same as test #4]
Test #18

4th order
Test #6 [treble clear]
Test #9
Test #12 [sounded like test #2]

5th order
Test #7 [must be test #2]
Test #10 [sounded like test #2]
Test #14

[Johnloudb]

Mark,

I heard a clear audible difference between these and I haven't done real critical listening yet. The unmodified one was best, of course .

I think the second order was pretty good, but I should listen more to that one.

The fourth order and upwards sounded worse.

That's about all can say for the moment. I listened on small 3" full range speakers driven by a T-amp. So, no crossover to mess with the phase. I've always found good 1st order speakers sound more natural to me.

[tf1216]

Mark,

Can you please check the link to the MATLAB code please? I was unable to download the files.

[---k---]

Chris,

That looks like a good way to do the test. Your results look like they show a little consistency. I'll have to try it this weekend.

[Davey]

Mark,

I didn't look at all the files, but there's something funny going on with the Alice in Chains selections. The unmodified file has been peak limited somehow at -6dbFS and looks visibly different than the processed ones. That should be easily audible.
A high degree of limiting is not unusual with much popular music, but that track is limited well below 0dbFS.

The average level of all these files should be consistent, but the peak levels will likely be different when the all-pass characteristic is applied.

Cheers,

Dave.

[Xander]

I will try this out with my Sennheiser HD280 Pro headphones this weekend...

A blind test of course.

Thanks for putting this up!

[Amphiprion]

Dangit Davey, you just caught another error!

All the songs had to be amplitude corrected. The 8th and 10th order filters would generate data outside the range of [-1,1] on all the songs. All the Natalie Merchant and Miles Davis tracks were amplitude modified by 0.8 and the Alice in Chains by 0.5. If I didn't do this I would get error messages about clipped data when trying to write the filtered results to a file. So brilliant me, in deciding to make the SPL the same for them all, multiplied the filtered results by the above numbers to normalize them all to the other versions of the same song.

Stupid thing is, a) I must have made a mistake with the Alice in Chains data and b) I should have reduced the amplitude BEFORE filtering, not afterward.

Nuts to me. OK, everybody disregard the music that has been posted so far. The SPL levels of the songs ARE NOT THE SAME. I have to go back and re-do the song processing. This is what happens when I try and get something done quickly... argh.

[Amphiprion]

Download links have been deleted until corrected wave files are posted

Apologies to those who have spent time listening. I'll make sure I do it right this time.

[Davey]

Mark,

I've been there. This kind of thing is tricky.

Cheers,

Dave.

[crackyflipside]

Disregard my tests, I was using it on speakers that have a crossover.

[Johnloudb]

Disregard my tests, I was using it on speakers that have a crossover.

Your testing method is certainly the way to go. Once Mark gets these files straightened out we can all give it another shot.