[safulop]

   
I believe this is what he's talking about: http://www.cerlsoundgroup.org/Kelly/soundmorphing.html
 
I'm not so sure how applicable it will be. The motivation of Kelly seems to make sounds with differing harmonics blend seamlessly together, but for our use I fear the extra processing might just be another confounding variable, and it should really be necessary anyway as the sounds we want to compare will have the harmonics all aligned. If they weren't, then we wouldn't need something more sensitive than regular ABX to tell them apart.

Yes, that's my occasional coauthor Kelly Fitz.  I was thinking perhaps the morphing could be used to make a seamless transition even though it is not needed to align the spectral components.

 

[safulop]

  There are several other avenues to pursue. Neuroimaging could prove useful: hook someone up to an fMRI during a randomized AB, whether soundmorphed or back to back. Establish thresholds for patterns of activity which occur when the test subject senses distortion. Unfortunately this approach is prohibitively expensive, but there's an increasing amount of neuroimaging studies which concern the perception of audio, or at least aspects of it (pitch discrimination, rhythm retrieval).
 
Alternately save yourself the trouble and see if there's a measurable difference beyond experimentally established thresholds.

Well we know there are measurable differences.  If you compare MP3 files from different encoders, same bit rate, at the fine structure of the waveform, they will not match up exactly when transcoded back to lossless.  But the remaining question is can anyone hear this difference that everyone can see in the analysis software?

 

[anetode]

  Well we know there are measurable differences.  If you compare MP3 files from different encoders, same bit rate, at the fine structure of the waveform, they will not match up exactly when transcoded back to lossless.  But the remaining question is can anyone hear this difference that everyone can see in the analysis software?

 
Yes, I should have specified that that isn't really the best approach for codecs. Although so far the psychoacoustic models employed in compression have been tested thoroughly and scored quite well. The question then is whether our understanding of psychoacoustics is sufficient for practical application and whether there is room for improvement. There has also been thorough testing of masking effects outside of codecs.

 

[Joe Bloggs]

Well, I could hear differences using ABX methods, but it just didn't seem as obvious as I thought it should.  So maybe I am mistaken, and maybe if I devise this whole "sound morphing" method or whatever it actually won't make any difference in the results.  But it will be interesting to check into it, if I end up going seriously into it.  In my research I have often focused on measurement methods, though in the past this has concerned mostly spectrum measurement, phase measurement etc.  I have not delved into perceptual psychophysics before, so this will be a steep learning curve.  But at least I know where I could drum up twenty subjects to do the listening

I once made a song file that morphed from flat to 10dB treble gain (with a gentle slope) in the middle of the song half way through, with a 3s transition period. Level matched by bringing the gain of the whole EQ down 5dB, of course. It was hard for someone to tell that anything had been done to the song, let alone point out the transition point. These things may just not be as obvious as your eyes on the EQ tell you they should be.

 

[safulop]

I once made a song file that morphed from flat to 10dB treble gain (with a gentle slope) in the middle of the song half way through, with a 3s transition period. Level matched by bringing the gain of the whole EQ down 5dB, of course. It was hard for someone to tell that anything had been done to the song, let alone point out the transition point. These things may just not be as obvious as your eyes on the EQ tell you they should be.

Hmm that's an interesting test.  But what if it was a 30 ms transition or a 3 ms transition, rather than taking 3 s for the effect?  I wonder if the transition point would be more clear.  Really, this is the job of psychophysics.  We have to do like 10,000 experiments and figure out all the audibility thresholds for all this type of stuff.  I realize that a lot has been done through ABX testing already, so not wanting to completely reinvent the wheel.

 

[arnyk]

  Hmm that's an interesting test.  But what if it was a 30 ms transition or a 3 ms transition, rather than taking 3 s for the effect?  I wonder if the transition point would be more clear.  Really, this is the job of psychophysics.  We have to do like 10,000 experiments and figure out all the audibility thresholds for all this type of stuff.  I realize that a lot has been done through ABX testing already, so not wanting to completely reinvent the wheel.

 
Or maybe you have to do only one test that shows that right in the middle of audibility it is totally inaudible, plus maybe a couple of more near the extremes. Inference works! Depends on what you are testing.

 

[prot]

... Depends on what you are testing.

That is a very important aspect which is quite often forgotten. One of the big advantages of ABX is that you can use it to test pretty much any kind of audio diffs: sample rate changes or any other bitlevel change in the audio file, changes in the analog domain or any two audio components (players, amp, dacs, etc).

I found safulop's color swatch analogy very interesting and the ABX method surely has some issues in the sensitivity area.
That morphing test also has some potential but you can only test changes in the digital domain ... one can of course test everything else by redigitizing the resulted analog wave but the extra ADC process comes with it's own sensitivity loss.

Here are a few other ideas for specific tests that may be more sensitive than abx:
1. For bitrate comparisons: mp3 already supports something called VBR and the fact that it works so well already proves that bitrate changes arent very audible. But one can contact the LAME devs and it should be possible to customize the convertor so that it changes the bitrate say 3x per song. And you can test if you can hear it.
2. (Almost) Any bit-change in an uncompressed file (e.g. wav files). It should be possible to cut a wave file at a very precise moment and than glue the pieces back together in an inaudible manner .. the file is just a sequence of bits and you know exactly where the cut was done (i.e. between bits xyz and xyz+1). You make two copies of one file, alter one in some way and than you splice the files and make a 'mix' by glueing pieces from the two files .. see if it's audible.
3. Test audio components. You can compare mono signal from component1 through left speaker and mono from component2 through right speaker. Or you can just use one channel of a stereo signal (e.g. left signal from component1 vs. left signal from component2)

While any of those tests *may* be more sensitive than ABX, you kinda lose the general applicability of ABX ... although no3 should be quite good in that respect. In any case, someone has to do the work and check if any of the above is indeed more sensitive than abx.

 

[arnyk]

That is a very important aspect which is quite often forgotten. One of the big advantages of ABX is that you can use it to test pretty much any kind of audio diffs: sample rate changes or any other bitlevel change in the audio file, changes in the analog domain or any two audio components (players, amp, dacs, etc).

I found @safulop's color swatch analogy very interesting and the ABX method surely has some issues in the sensitivity area.
That morphing test also has some potential but you can only test changes in the digital domain ... one can of course test everything else by redigitizing the resulted analog wave but the extra ADC process comes with it's own sensitivity loss.
 

 
The idea that a test involving redigitizing necessarily causes a sensitivity loss is a just hypothesis, not a global fact. It is easy to falsify by several means.
 
(1) The technical performance of really good digitial converters is so good that knowing what we know about human hearing, there should be no audible loss.
 
(2) Any number of listening tests have been done involving digitizing and redigitizing recordings, and no audible differences have been found if the converters are really good.

 

[StanD]

   
The idea that a test involving redigitizing necessarily causes a sensitivity loss is a just hypothesis, not a global fact. It is easy to falsify by several means.
 
(1) The technical performance of really good digitial converters is so good that knowing what we know about human hearing, there should be no audible loss.
 
(2) Any number of listening tests have been done involving digitizing and redigitizing recordings, and no audible differences have been found if the converters are really good.

Even though this is true, IMO one should not make any such assumptions when testing.

 

[arnyk]

  Even though this is true, IMO one should not make any such assumptions when testing.

 
Please show us a real world test that was completely free of assumptions.

 

[StanD]

   
Please show us a real world test that was completely free of assumptions.

That's not an excuse to go to the max.

 

[arnyk]

  That's not an excuse to go to the max.

 
The criticism was precisely stated as: 
 
"Even though this is true, IMO one should not make any such assumptions when testing."
 
Key words: "Any such assumptions." So, now we have an imposed order where some assumptions may be OK but others are to be avoided at all costs.
 
Here are the assumptions that were proposed to banned outright:
 
The idea that a test involving redigitizing necessarily causes a sensitivity loss is just a hypothesis, not a global fact. It is easy to falsify by several means.
 
(1) The technical performance of really good digital converters is so good that knowing what we know about human hearing, there should be no audible loss.
 
(2) Any number of listening tests have been done involving digitizing and redigitizing recordings, and no audible differences have been found if the converters are really good.
 
Yes, those are both extant scientific findings. Why be so eager to supress them?  Do they cause discomfort?  Are scientific findings that cause discomfort to be supressed?
 
Note that both means have been validated scientifically, yet an assertion has been made that they are to be avoided at all costs.
 
The key words here are IMO: "Validated Scientifically". I see an attempt to cherry pick the Scientific findings that are to be allowed.
 
Is that what happened?
 
Why?

 

[StanD]

   
The criticism was precisely stated as: 
 
"Even though this is true, IMO one should not make any such assumptions when testing."
 
Key words: "Any such assumptions." So, now we have an imposed order where some assumptions may be OK but others are to be avoided at all costs.
 
Here are the assumptions that were proposed to banned outright:
 
The idea that a test involving redigitizing necessarily causes a sensitivity loss is just a hypothesis, not a global fact. It is easy to falsify by several means.
 
(1) The technical performance of really good digital converters is so good that knowing what we know about human hearing, there should be no audible loss.
 
(2) Any number of listening tests have been done involving digitizing and redigitizing recordings, and no audible differences have been found if the converters are really good.
 
Yes, those are both extant scientific findings. Why be so eager to supress them?  Do they cause discomfort?  Are scientific findings that cause discomfort to be supressed?
 
Note that both means have been validated scientifically, yet an assertion has been made that they are to be avoided at all costs.
 
The key words here are IMO: "Validated Scientifically". I see an attempt to cherry pick the Scientific findings that are to be allowed.
 
Is that what happened?
 
Why?

If one is testing to determine if any differences can be heard, then any unnecessary alterations should avoided. Simple idea, one can argue that the alterations interfered with the tests, you can be certain that those who worship hirez will use that as an argument.

 

[prot]

The idea that a test involving redigitizing necessarily causes a sensitivity loss is a just hypothesis, not a global fact.

True. It's just better not to introduce extra variables and give people reasons to complain about. I dont need to tell you how much energy was spent talking about how the ABX boxes make everything sound the same...

 

[nick_charles]

  If one is testing to determine if any differences can be heard, then any unnecessary alterations should avoided. Simple idea, one can argue that the alterations interfered with the tests, you can be certain that those who worship hirez will use that as an argument.

 
From my observations here and elsewhere high rez proponents do not need rational criticisms they have a vast arsenal of attacks such as (all found , none made up)
 
 
1) You must be deaf if you cannot hear a difference
2) Your kit is not good enough if you cannot hear a difference
3) I don't need to submit to a blind test, I know I can hear a difference
4) The source material was not good enough to be able to hear a difference (sometimes this argument may be justified see Meyer and Moran)
5) You are not listening seriously enough to hear a difference
6) 16 bits cannot properly express any real world analog sound
7) Just because scientifically we can't hear something doesn't necessarily mean we don't perceive it
8) Music is not sine waves
9) High res provides more nuances, transparency ambience and realism
10) There is something missing from red book
11) higher frequency content isn't to capture nominal or "carrier" frequencies above 24khz, but to capture the small fluctuations on frequencies in the audible range
12) 16 bit 44.1-khz sampling "absolutely demolishes any part of the signal above 10k"
13) DSD introduces a sound signature that PCM does not inherently have, and this sound signature does not lend itself to digital down conversion to 44.1khz/16bit PCM; but it somehow is able to come through if DSD is converted to analog first.
14) I can hear a  difference
15) There is obviously a difference
16) I can easily tell the difference between CD version and highres version of the same remaster. It is simply sounds more sophisticated and less harsh.  
17) I can hear a difference 
18) there exist audible signals that cannot be encoded transparently by a standard CD\
19) I can hear a difference