Thanks Mark, I had never heard of that one. I still keep thinking I can hear a difference at times and yet the ABX results often say I can't.
The other throw of the dice is that listening to one at a time (or a select part of one) and then taking a few seconds break as I load another track into media player seems for me to be the way that I think I hear the differences. And yet carrying those differences over to an ABX is not working. Hmmm
The other throw of the dice is that listening to one at a time (or a select part of one) and then taking a few seconds break as I load another track into media player seems for me to be the way that I think I hear the differences. And yet carrying those differences over to an ABX is not working. Hmmm
One thing I do seem to notice more with sighted listening is a tendency to focus on different aspects of sound on the different tracks. It seems possible to overcome that though, by intentionally listening for aspects on one track believed to be associated with another track. Once that bias is compensated for, the differences seem more subtle.
However, I would kind of expect someone with young yet well trained ears listening on a very accurate system would have the best chance of noticing any differences. You have younger ears than me by about a decade, and probably better headphone transducers, but possibly I have a little better DAC. If we could get the best of everything together at one time and in one place, it might be very interesting.
15 years ago or so, I could hear added distortion from almost any digital EQ, except for one or two of the very best upsampling ones. But my ears could still hear HF well then, and I was in probably in peak ear training shape, practicing every day.
The above considerations make me suspect there must be somebody around we could find who could be used to get a pretty good idea of the best people can do under the most ideal circumstances. If somebody wants to come over here and try, I'm maybe around an hour north of Sacramento.
However, I would kind of expect someone with young yet well trained ears listening on a very accurate system would have the best chance of noticing any differences. You have younger ears than me by about a decade, and probably better headphone transducers, but possibly I have a little better DAC. If we could get the best of everything together at one time and in one place, it might be very interesting.
15 years ago or so, I could hear added distortion from almost any digital EQ, except for one or two of the very best upsampling ones. But my ears could still hear HF well then, and I was in probably in peak ear training shape, practicing every day.
The above considerations make me suspect there must be somebody around we could find who could be used to get a pretty good idea of the best people can do under the most ideal circumstances. If somebody wants to come over here and try, I'm maybe around an hour north of Sacramento.
I think that digital has improved a lot in the last 15 years. Before about 2004, most CD players were a c..p, regarding sound. Now, we have very good DACs and can play high resolution files from PC or notebook. USB and Toslink jitter has been solved. I am sending digital through optical SPDIF to the DAC, so it is completely free of interferences. And the sound quality is almost perfect, then.
I dont think you can use this method when the difference is subtle. The best method is to group several subtle different notes into one small chunk and familiarize your ears with the sound of this chunk. Easier with string instruments, which might not be much in the clip.
Another method is just the "practical implementation" of the above in real listening. We listen to all the clips in random and when we think we like the clip we peek and note the clip. Same when we dont like a clip we also peek and note it...
By doing the above, we are comparing the "timing" or correct rythm. I'll guess the TL072 and the 1458 will get the most "OK" votes with 1458 has more "NG" votes between them. OPA2134 will have the most "NG" votes and LM1562 will have the least votes (i call that boring with no soul).
If doing twenty 8 trial ABX-tests at SL=0.05 (actual SL is 0.035), then the probability to get at least 1 success by chance is p=0.509 , so quite likely.
But doing an ABX consisting of a small number of trials is only recommended if the listener´s detection ability is really good, otherwise the risk of commiting a type 2 error gets unbearable.
Thanks Mark, I had never heard of that one. I still keep thinking I can hear a difference at times and yet the ABX results often say I can't.
The other throw of the dice is that listening to one at a time (or a select part of one) and then taking a few seconds break as I load another track into media player seems for me to be the way that I think I hear the differences. And yet carrying those differences over to an ABX is not working. Hmmm
It could simply be that the specific ABX protocol does not fits you individual mode very well. ABX test involves different internal mental processes in comparison to A/B tests.
Unfortunately i don´t know if a common software exists for doing pure A/B comparisons (means preference paired comparison), but maybe to switch over to an ABC/HR could also help.
Yes, this is sometimes mentioned, that the ABX test involves different mental processes than the AB test. My knowledge of psychoacoustics is not that good to confirm it or deny it.
I can only say that I was trying hard to prepare good "analog path" conditions for the test and that the test circuit (AB box mentioned previous, with Alps and unity gain buffer) is the most transparent preamp I have at the moment and it has both LF and EMI interferences completely fixed. This is the key and believe me or not, differences between sound of the op amps start to wipe away then.
Experimenters did several comparisons between A/B- tests and ABX versions (somewhere around 1954-1956 when the original ABX test protocal was introduced) and noticed that the participants did better in the A/B than in the ABX trials. They attributed the difference to the internal mental processes, although i have to reread some publications to recollect if they further tried to confirm this hypothesis.
Intuively the processes must be different, because when following the ABX protocal participants have to memorize and process three stimuli compared to only two in the A/B test.
I would agree (at least that it is a valid and plausible hypothesis) .....
To small number of trials can get type 1 errors.
To eliminate type 2 errors, just take your time, take pauses, there's no limit on how much time you should spend. And then you can do 20 trials eliminating both type 1 and type 2 errors.
For preference tests with very small differences, like this test, ABC/HR should be used. BS.1116 : Methods for the subjective assessment of small impairments in audio systems
I'm more perturbed by the fact that I can't immediately pick what is essentially a 741 from a cluster of better devices. Not quite sure what that is telling me actually......
My current method is to play either the opening or to jump in at a certain point but not to ABX them. I then score them out of 10 giving a middle score of 5 to the first device (simply because it is first in the list) and then marking the others as either same, higher or lower.
When I think I have found a preference for one or more I then put that plus the worst scoring into the ABX procedure and become despondent that the differences melt away.
My current method is to play either the opening or to jump in at a certain point but not to ABX them. I then score them out of 10 giving a middle score of 5 to the first device (simply because it is first in the list) and then marking the others as either same, higher or lower.
When I think I have found a preference for one or more I then put that plus the worst scoring into the ABX procedure and become despondent
that the differences melt away.in AB/X listening test, to me a pairwise comparison is a 'unit'
I mainly listen pairwise in AB/X, A vs B to try to find, fix 'the difference" in my memory then alternate A vs X and X vs B, with A vs B, frequently 'refreshing' listens to A vs B
even when when something seems 'obvious' I test A vs X, B vs X is 'same or different?' rather than just trying X and calling it as A or B
I mainly listen pairwise in AB/X, A vs B to try to find, fix 'the difference" in my memory then alternate A vs X and X vs B, with A vs B, frequently 'refreshing' listens to A vs B
even when when something seems 'obvious' I test A vs X, B vs X is 'same or different?' rather than just trying X and calling it as A or B
An experimenter chooses the significance level at which he wanted to test at first and from there he gets the minimum number of trials needed.
Elimination of both error types isn´t possible, just getting both error risks at the same level (presumably).
While accomodation to the test protocol will probably help a lot, without positive controls the experimenter doesn´t know about the risk of commiting an error of the second kind.
He has to assume several detection probabilities and to do a power calculation under these assumptions.
As said before, ABC/HR is a nice protocol but A/B paired preference comparison well planned and executed is another one.
When do we get to know what the chips are ?
I used a Philips SHP8900 headphone via my sound card which is quite decent.
Tomorrow I will try it again using my Beyer DT990. Maybe play the files via my external DAC / headphone amp.
The differences on the SHP8900 aren't huge but some parts seem distinctly different. Will have to do it all over again and see if I have the same ratings I made in the first round on the SHP8900.
I used a Philips SHP8900 headphone via my sound card which is quite decent.
Tomorrow I will try it again using my Beyer DT990. Maybe play the files via my external DAC / headphone amp.
The differences on the SHP8900 aren't huge but some parts seem distinctly different. Will have to do it all over again and see if I have the same ratings I made in the first round on the SHP8900.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.