Hires 96/24 listening test of opamps

[PMA]

As a follow-up to the thread
http://www.diyaudio.com/forums/everything-else/308514-another-big-opamp-listening-test.html
I am posting a new opamp test, this time in 96kHz/24bit resolution. The music samples may be downloaded from
http://pmacura.cz/test_hires.zip
The file is quite big, about 160MB, as a result of higher resolution format.

The samples have a lot of information above the audio band, as you can see from the attached spectrogram. It is a true hires, not the file create by ultrasonic noise addition, as is sometimes quite usual.

The link level was now 2Vrms = 0dBFS.

The poll is valid for 10 days. Please do not write your comments to the forum, but rather send it to me to my PM here or to my e-mail address.

It would be nice if you could support your preferences by an ABX result, should be signed like in this example:
http://www.diyaudio.com/forums/ever...-big-opamp-listening-test-25.html#post5109759
This allows verification of validity of your result.

Thank you and have fun!
--------------------------------

Poll question "I can not hear a difference" was added on June 25. Please feel free to vote for this option in case that all the files sound same to you, without audible difference.
----------------------------------
June 27
The poll has been closed and it's time to reveal the parts used:

rr – LM4562 (1 vote)
ss – OPA2134 first take (2 votes)
tt – MA1458 (2 votes)
uu – TL072 (9 votes – big winner 🙂)
vv – OPA2134 second take (1 vote)

So, according to the poll, the winner is TL072 😀. I have also received preferences via PM and I tried to create a chart from them.

This shows a different view, very contradictory and there are not many mutual preferences among the voters.

Moreover, none of the votes was supported by a valid ABX result, so I would leave to a reader to make his conclusion himself. However, there is one very interesting result, “Voter 2” preferred the parts just according their parameters with one exception, two takes of OPA2134. Great job, Mark! 🙂

I have prepared some more information, together with measurements, and placed it to my web:

Op amps listening test with hires files

Thanks to everyone who participated in this test.

As a bonus, I am adding the original source file which was used in the test:
http://pmacura.cz/ww.zip

 

[Markw4]

Just to be clear Pavel, I think you said,

Please do not write your comments to the forum,

but rather send it to Pavel by PM as per the post above, or to his e-mail address.

 

[PMA]

Just to be clear Pavel, I think you said,

Please do not write your comments to the forum,

but rather send it to Pavel by PM as per the post above, or to his e-mail address.

Mark, you are absolutely right.

 

[scott wurcer]

Pavel, a full round would be 32 combinations is this necessary? I'm thinking of writing a Python script to randomly make up ABX combos to write to my SD card so I can do undisturbed headphone listening.

 

[PMA]

Pavel, a full round would be 32 combinations is this necessary? I'm thinking of writing a Python script to randomly make up ABX combos to write to my SD card so I can do undisturbed headphone listening.

Scott, it is definitely not necessary. For ABX, I would be glad if you tried just 2 pairs, any of the set.

 

[Mooly]

Would you like ABX results posted here Pavel ?

 

[PMA]

Would you like ABX results posted here Pavel ?

I think it would be fine, Karl.

 

[Mooly]

OK 😉 I think first of all I will just listen and make notes for myself rather than ABX (at least to begin within).

 

[PMA]

On a technical note: the worst opamp in this test has now THD at 10kHz = 0.064% and is about 20x worse in distortion at 10kHz than the digital chain. This is the result of higher voltage level in this test. Also, the differences in signals are higher than in the previous test. The best opamps again make a direct copy of the digital chain properties.

 

[Mooly]

I've just been listening and in the end did ABX a couple of pairs. Nothing conclusive yet 🙁 One for tomorrow and the coming days me-thinks.

 

[PMA]

Thank you, Karl. We have enough time 😉.

 

[PMA]

Well, I hope you guys will be much better than me in the ABX 😉

foo_abx 2.0.2 report
foobar2000 v1.3.7
2017-06-18 09:52:50

File A: rr.wav
SHA1: e374d11099f329141cf3cf553f646f49102aaaba
File B: tt.wav
SHA1: 1a956970e62c3aaeea483c29334833c82a612e84

Output:
WASAPI (push) : SPDIF-Out (USB Sound Blaster HD), 24-bit
Crossfading: NO

09:52:50 : Test started.
09:54:37 : 01/01
09:54:48 : 02/02
09:55:08 : 02/03
09:55:20 : 02/04
09:55:36 : 03/05
09:55:47 : 03/06
09:55:58 : 04/07
09:56:14 : 05/08
09:56:14 : Test finished.

 ---------- 
Total: 5/8
Probability that you were guessing: 36.3%

 -- signature -- 
1c7e442a5397bf8bfbd8170d85a1ba7cba58aca2

 

[Mooly]

Nope 🙂 I've had 3 goes at this and the best was 6/8. Twice last night and once today.

There is one file I keep thinking is the 1458, the other sound very similar.

More listening required.

 

[milkshake]

May I note that ALL trials should be included in the test results.

If you do for instance 20x8 ABX tests there's a very high probability that one of those 20 tests gives a 7 out of 8 right score just by chance alone.

 

[PMA]

There is one file I keep thinking is the 1458, the other sound very similar.

.

OK, so please keep it in secret 😀
The opamps are the same as in the previous test + one is added.

 

[Mooly]

Of course 🙂

I need to do more listening to these, in particular to try and see if I still feel the 'suspect' track really could be the 1458.

 

[PMA]

May I note that ALL trials should be included in the test results.

If you do for instance 20x8 ABX tests there's a very high probability that one of those 20 tests gives a 7 out of 8 right score just by chance alone.

That's correct. I kept 4/8 - 5/8 in 5 trials, that means just guessing 😀

 

[maty tinman]

Pavel, as I told you by mail, now it has taken me more to differentiate between them. I suppose it is due to the 2 Vrms and the greater separation from the noise.

And it cost me less work with the AV Marantz SR4500 (KEF Q100 speakers and the Takstar ts671 headphones, cheap but with very good sound after tweaked + recabling) than with the ifi iCAN even though the headphone amplifier has a SNR of 117 dB. Perhaps the iCAN's switched power supply is the bottleneck.

​

 

[Mooly]

Not doing very well here. The one I don't like seems to vary. The intro is good, vocals less so. I'm going to stick with that as the 1458 at the moment.

I've blanked out the tracks to keep the files that I'm choosing anonymous.

Maybe I should stock up on 741's and 1458's. After all, if you can't tell the difference then using the lowest cost device to achieve the desired result is called 'engineering' 😉

Who said that 😱

 

[Markw4]

Mooly, What I do to find a good spot to listen to is load all the files in Reaper, then un-mute one at a time, and go up and down though the tracks trying to notice when things sound a little different.

For example, right now I am listening to an interesting cymbal hit that is looping from 0:07.50 to 0:08.15. A little more than 1/2 second in the loop.

That particular hit has a little HF ring in it that I think may sound a little more or less grainy on some of the tracks. Wish I could still hear above 10 kHz, as I would expect most of the difference to be up there for this kind of test.

By the way, trying to continuously loop 1/2 second in Foobar while listening carefully is almost impossible for me. Maybe I'll ask someone to be the computer operator so I can give all my attention to listening.