Thanks for the explanation. I was just about to rename (with the help from a friend) earlier files and listen again. Good that you posted 'Reruns'. With quick listening of short clips of all the files yesterday and today, I liked violet. As all sounded similar I could be right or wrong. Either way I would be happy. Probability is 12.5 % without listening.The LM883P (it must be that version) uses a quasi complementary output stage (two NPN transistors instead of the usual NPN/PNP pair) which is reputed to have a more musical sound, particularly when used in power amplifiers.
If you could pick it out again, then that would be beginning to look like something really was identifiable
Regards.
If you run something like this again, throw a few controls in. For example, record a track twice with the same op-amps and see if they score the same in the rankings. Better still, do that with two or three of the opamps, then scatter them randomly. Make sure to erase or randomize the file time stamps, of course.
Sure this is one approach, but do we really need to get this anal for the purposes of this test.
IMO another worthwhile test methodology for this level/kind of testing is to in the first case identify the opamp used for each audio file and solicit subjective opinions.
Sure, this test may have results skewed by preconceptions caused by knowledge of internet 'opinions', but in this kind of 'informal' testing real results can be obtained.
In the listener opinion/findings sample set presented so far there is already revealed differences in system quality/resolution and differences in listening skill levels.....these are actually valid data points.
Commonality of subjective opinions of known particular opamp files as presented is quite valuable despite lack of DBLT protocols.
Mooly has gone to the trouble of setting up his DUT jig, and this mode of testing gives the opportunity of taking a listen to a range of opamps without financial outlay...kudos.
This kind of testing can perfectly easily progress to 'crowd sourcing/suggestion' model and become a very useful tool.
Dan.
That would be great in theory, in practice you would find all those with reputations at stake would never give complimentary comments about anything other than the current state of the art devices. In fact the reality is that those folk would not even take part... ooops I just picked xyz as being the best
SY's comments are inline with my initial request for multiple normals scattered in. Both should work well (all under the assumption of a well-behaved testee, which is pretty much not a guarantee), and avoid such issues as the ones Dan so wonderfully pontificates in his posts on this thread.
Haven't had a chance to do round two, hopefully soon.
Haven't had a chance to do round two, hopefully soon.
Exclusive VALIDITY of randomized testing
100% with you, Sy. Having done a bunch of very careful testing in my years (having put together a unique device-under-test ABC/X jig), I have been left with one and only one testing methodology: neither the listener, nor the tester should be aware of the selection of tested device / cable / etc.
My ABC/X box was controlled by a serial cable (in the old days, before USB); the computer, having smarts, could easily randomize the selection of all sorts of things. Source, destination, path. There were trimming potentiometers on all sources, so that with some painful setup, all levels could be normalized. (If you din't normalize levels, then people would almost inevitably choose the slightly louder A/B/C setup, especially as testing fatigue set in)
AND … just to add to the mess … I added "disinformation". The actual relays used to select signal path were remarkably quiet mercury-wetted silver types. The box internally though had dummy relays naturally based on placement and box reverberation, with subtly different sounds. Very acute testers immediately learned those click-differences. They weren't aware that the computer was triggering relays completely independent of the tested-device selection. LOL!
The user (tester) had a box that allowed testing of A, or B, or C (3 source) or RANDOM. A, B & C were however also tied to a particular noise-generating relay. Only RANDOM was both randomized by ABC selection and noise generating relay. Because it was the analog days, I put 3 "qualities" large knobs on the top of the box, to record the tester's feelings about things. Using AUDIO pots and a logarithm scaled voting scale, they could listen, turn knobs, and press the "judge" button, resetting the thing. Very analog, but also much loved by those who participated. EASY to use.
This was somewhat before PCs were outfitted with audio cards (and had the hard disk space) to record tracks, such as per this thread's test methodology. But there were 8 inputs (L+R) x 4 plus a 2 input-to–8 output buffer (allowing "inbetween testing"). You could plug one source into the buffer, then split it 4 ways to 4 test devices (one of which was a straight-thru cable, for the Control), then collect back the 3–4 outputs into the ABC/X channels, which would then get switched onto the pair of outputs.
And you had knobs!
Anyway, back at you, Sy. ABC/X plus randomization and source anonymity is vitally important if the results are to mean much of anything.
GoatGuy
100% with you, Sy. Having done a bunch of very careful testing in my years (having put together a unique device-under-test ABC/X jig), I have been left with one and only one testing methodology: neither the listener, nor the tester should be aware of the selection of tested device / cable / etc.
My ABC/X box was controlled by a serial cable (in the old days, before USB); the computer, having smarts, could easily randomize the selection of all sorts of things. Source, destination, path. There were trimming potentiometers on all sources, so that with some painful setup, all levels could be normalized. (If you din't normalize levels, then people would almost inevitably choose the slightly louder A/B/C setup, especially as testing fatigue set in)
AND … just to add to the mess … I added "disinformation". The actual relays used to select signal path were remarkably quiet mercury-wetted silver types. The box internally though had dummy relays naturally based on placement and box reverberation, with subtly different sounds. Very acute testers immediately learned those click-differences. They weren't aware that the computer was triggering relays completely independent of the tested-device selection. LOL!
The user (tester) had a box that allowed testing of A, or B, or C (3 source) or RANDOM. A, B & C were however also tied to a particular noise-generating relay. Only RANDOM was both randomized by ABC selection and noise generating relay. Because it was the analog days, I put 3 "qualities" large knobs on the top of the box, to record the tester's feelings about things. Using AUDIO pots and a logarithm scaled voting scale, they could listen, turn knobs, and press the "judge" button, resetting the thing. Very analog, but also much loved by those who participated. EASY to use.
This was somewhat before PCs were outfitted with audio cards (and had the hard disk space) to record tracks, such as per this thread's test methodology. But there were 8 inputs (L+R) x 4 plus a 2 input-to–8 output buffer (allowing "inbetween testing"). You could plug one source into the buffer, then split it 4 ways to 4 test devices (one of which was a straight-thru cable, for the Control), then collect back the 3–4 outputs into the ABC/X channels, which would then get switched onto the pair of outputs.
And you had knobs!
Anyway, back at you, Sy. ABC/X plus randomization and source anonymity is vitally important if the results are to mean much of anything.
GoatGuy
Its all gone quiet
Yes, maybe the thread has run it course, but it has been very informative and a fun learning experience. I am grateful for the time and effort you put into making this work.
What I take away from it is that on speakers, I could not differentiate between any of the op amps, but on a pair of Stax headphone I could, barely. Since 5 op amps in series is about what happens in the active crossovers I make, this is relevant information to me. By using a solid and affordable performer like the NE5532, you can't go wrong.
Oh, wait, we have not yet finished learning!!!
Max Headroom informed us that he could hear pronounced differences between the different op amps, and so you put together the test again, with some improvements. We haven't heard from him since. The question if some people can really differentiate between op amps, is even more relevant to me than if I can do so. I don't make my speakers to keep them to myself.
This being said, great thread and thanks again.
I wasn't able to make time to listen to the files as planned over the weekend, sorry about that.
I will endeavor to run the test in the next couple of days and give subjective opinion of what I find.
Dan.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.