Organic yellow mustard, culinary mp3. In a squeezy weezy bottle too!
Buy Yellow Mustard, Organic,16 ozs. | Health Food Stores | Organic Food
I found it pretty easy - at least my ears told me so - to group the files, but neither group sounds like my original (from the Katie Melua Collection), which is slightly softer and less loud. So besides the allpass processing also the resampling (and fading and low level processing like maybe some dithering) made a significant difference in my opinion. The distorted parts are the same of the original and the files of this test.
That's how it goes... whenever there is a chance to prove that their ears really are golden they act almost a bit insulted by the opportunity.
Thanks- I eventually got them to him. If anyone else wants renumbered files, drop me a pm and I can do what Stein suggests.
That's how it goes... whenever there is a chance to prove that their ears really are golden they act almost a bit insulted by the opportunity.
I, for one, don't find any need to prove anyone anything concerning my earing.
SY and anyone else interested
You may have a look here
http://www.diyaudio.com/forums/software-tools/195457-spreadsheet-can-do-spectral-magnitude-phase.html#post2688368
Regards
George
As for the test files. After I learned that they were all pass filtered, I tried some of that on my own. I took a file, low pass filtered it 4th order at 1Khz and set that aside. High pass filtered at 1Khz and set that aside. Then mixed the HP and LP together. The recombined file sounds remarkably like the original, except there seems to be a tiny bit of extra reverb or ambiance to the filtered file.
I actually kinda like it.
I actually kinda like it.
@Jannemans phase twisting test:
Initially, I mailed a null result to Jan. SY then was kind enough to point met to Foobar ABX, which I installed and tried this weekend. Very interesting.
My initial listening tests where done in open listening, playing files one after another. It was for me not possible to register differences with any degree of confidence, so I informed Jan about a null result and later posted that conclusion.
However, with Foobar ABX, things are not the same. I scored 7/7 (> 99% confidence) in the first run.
Now, the question is why, and my answer lays in the transitions between the different files while switching. Identical files may have some sort of a click, sometimes, but that is about it.
However, switching between files with and without the phase shift gives a very different effect, with wierd things like twisted echos or reversed cymbals occuring briefly. But, after this switching artifact is gone, I couldn't tell for the world the two different files apart. I ABX-t again while muting during switches, and I can't hear a darned difference again.
For me, this is not so much about how good my ears are - they serve me well - but about what the limits of human perception are. What I think this experiment shows to me, is that I am able to register a sudden phase shift. My brain seems to get tuned to a certain set of phase relationships, and if these change rapidly, it gets noted. However, even with Foobar, I could not distinguish the files with muting the transition period. In other words, absolute phase doesn't register (with me).
I am just curious to find out what is behind it, so I would like to learn from Pano and SY if they can keep the files apart, blanking out the moment of transition.
vac
Initially, I mailed a null result to Jan. SY then was kind enough to point met to Foobar ABX, which I installed and tried this weekend. Very interesting.
My initial listening tests where done in open listening, playing files one after another. It was for me not possible to register differences with any degree of confidence, so I informed Jan about a null result and later posted that conclusion.
However, with Foobar ABX, things are not the same. I scored 7/7 (> 99% confidence) in the first run.
Now, the question is why, and my answer lays in the transitions between the different files while switching. Identical files may have some sort of a click, sometimes, but that is about it.
However, switching between files with and without the phase shift gives a very different effect, with wierd things like twisted echos or reversed cymbals occuring briefly. But, after this switching artifact is gone, I couldn't tell for the world the two different files apart. I ABX-t again while muting during switches, and I can't hear a darned difference again.
For me, this is not so much about how good my ears are - they serve me well - but about what the limits of human perception are. What I think this experiment shows to me, is that I am able to register a sudden phase shift. My brain seems to get tuned to a certain set of phase relationships, and if these change rapidly, it gets noted. However, even with Foobar, I could not distinguish the files with muting the transition period. In other words, absolute phase doesn't register (with me).
I am just curious to find out what is behind it, so I would like to learn from Pano and SY if they can keep the files apart, blanking out the moment of transition.
vac
Last edited:
As for the test files. After I learned that they were all pass filtered, I tried some of that on my own. I took a file, low pass filtered it 4th order at 1Khz and set that aside. High pass filtered at 1Khz and set that aside. Then mixed the HP and LP together. The recombined file sounds remarkably like the original, except there seems to be a tiny bit of extra reverb or ambiance to the filtered file.
I actually kinda like it.
Pano is that 'reverb'- like effect similar to what you hear in the x1000 file?
jan didden
That's exactly the point- this isn't a test of listeners, it's a test of perception. My experience wasn't too different than yours- listening to the files sequentially didn't do much for me, but when I started switching rapidly, the differences became more evident. I don't know how to put a dead space into the switching (software is not my gift), but I suspect that would reduce sensitivity (at least my sensitivity) to hearing the phase shift.
I'm still surprised (and to be honest, somewhat disappointed) that so many people were militantly incurious. This sort of comparison is fascinating, and I thank Prof. Hawksford and Jan for putting this listening test out there.
No, not really. The 1000X file has awful tweeting and warbling. Now maybe it is the same thing I hear at a low level, but subjectively it's very different at those levels. So I don't hear a correlation between the two at all. It surprising.
Both groups of files are seemingly processed - some kind of resampling, processing via filters or passing through transformers (all very subtile but audible). Any stage of computer processing creates - if you like it or not - a new file which is slightly different to the former: But that is what we do in mastering any day.
For me, working mainly at the creation side of audio recording and mastering, these kind of differences are quite common, and are not discussed as in good or bad, but as possible versions or production/mastering decisions, done on purpose. There is no problem about these differences if they are that small and if there is no apparent reason to find one better than the other or not.
Thank you for the test.
One more remark: It is also a very problematic test indeed. The original file (I suspect in the "minor" 16bit/44.1kHz integer format) is, for a process like this one, resampled in some higher format (maybe 32bit/44.1khz floating point, the bit-depth in floats is important here) to do the calculation of the filtering (hopefully done in double precision float). The results are then resampled to 16bit/44.1kHz again. This process is per default a lossy one, because the truncation of the bit-depth is necessary. This is usually done with some kind of tricks to make it sound good (a simple truncation never sounds good enough), some rounding and dithering for example. Therefore even if any calculation was done without significant influence, the resulting file is slightly different to the original. That may be the reason why my original still sound different to all the test files - even the editing to different lenghth and the in and out fading process is definitely altering the file to something new and different.
For me, working mainly at the creation side of audio recording and mastering, these kind of differences are quite common, and are not discussed as in good or bad, but as possible versions or production/mastering decisions, done on purpose. There is no problem about these differences if they are that small and if there is no apparent reason to find one better than the other or not.
Thank you for the test.
One more remark: It is also a very problematic test indeed. The original file (I suspect in the "minor" 16bit/44.1kHz integer format) is, for a process like this one, resampled in some higher format (maybe 32bit/44.1khz floating point, the bit-depth in floats is important here) to do the calculation of the filtering (hopefully done in double precision float). The results are then resampled to 16bit/44.1kHz again. This process is per default a lossy one, because the truncation of the bit-depth is necessary. This is usually done with some kind of tricks to make it sound good (a simple truncation never sounds good enough), some rounding and dithering for example. Therefore even if any calculation was done without significant influence, the resulting file is slightly different to the original. That may be the reason why my original still sound different to all the test files - even the editing to different lenghth and the in and out fading process is definitely altering the file to something new and different.
So here:
'the aritmetic had a dynamic range exceeding 250 dB, it was then correctly dithered and truncated. this could have been to 24 bit but then the files would be easily identified and maybe not playable on all systems. obviously all processing offers loss, even taking a file and just dithering and quantizing. why not try comparing the phase between the 2 tracks. so yes requantizing to 16 bit may by itself maybe detectable... i should have requantized the non phase distorted track.'
jan
'the aritmetic had a dynamic range exceeding 250 dB, it was then correctly dithered and truncated. this could have been to 24 bit but then the files would be easily identified and maybe not playable on all systems. obviously all processing offers loss, even taking a file and just dithering and quantizing. why not try comparing the phase between the 2 tracks. so yes requantizing to 16 bit may by itself maybe detectable... i should have requantized the non phase distorted track.'
jan
- Status
- Not open for further replies.