Can you hear a difference between OPA2134 and LME49720 in the PRE-TC10?

[antcollinet]

There are some people who undertook the ABX test and the result show that they can hear a difference; which I find interesting.

Unfortunately 8 trials are not sufficient, 10 is the recommended minimum.

ABX test - Wikipedia

en.wikipedia.org

9/10 is considered "statistically significant" but is still not absolute proof: there is still a residual (around) 5% chance that it was pure guessing.

If you want something close to proof of audibility, you need to be targeting a p value of 0.01 (1%), or even lower.

 

[DanielT]

Thank you, Klaus. In fact I have more repeated runs (4) and they all say the same - the difference between 2134 and 49720 with the test music sample is consistent and is not the error of subsequent recordings. It takes time, you know. The differences strongly depend on music samples used. Woman voice sibilants with HF transients are sensitive to subtle HW differences. On the other hand, Tracy Chapman’s Fast Car gives me no potentially audible differences. I do not know why this track is used for ABX tests. Maybe it is fine for speaker testing (though I doubt it) but definitely not for the tiny op-amp differences.

Maybe Sierra Hull has more of the Woman voice sibilants with HF transients that you are looking for? There are tons of female singers, but I took Sierra Hull just because I think she is similar to Tracy Chapman in style. With the difference that Sierra Hull has an Appalachian style vs. that Tracy Chapman's feels more city.

I don't know if there are good dynamics in the recording:

Edit:
Also, a lot of banjo and violin in that album. And with that energy in the higher frequency registers, which I understood what you were looking for.
You can listen to it yourself and decide.

 

[restorer-john]

Find some really good jazz recordings made on Soundstream digital gear in the early 80s with the 50kHz SR where they (I think) used the early Studer SRC for 44.1 CD, ones with some cymbals, bell-trees or chime-bar instruments well recorded. HF distortion/harmonic truncation(?) in my experience manifests as a "mush" instead of detail.

It's always a giveaway with compressed audio.

Ask @amirm His jangling keys are a classic and I'm sure they/he had some particular tracks that would show/hide good/poor codecs. Same could be used for this test.

@pma Did you answer the bandwidth question on your THD vs Freq plot?

 

[DaneelOlivaw]

???
That's exactly what this test is, you have no way of knowing with foobar's ABX test, that's its whole point.

I doubt you got the point. The ABX test is a precision instrument. Just like with a microscope, the smallest differences can be analyzed. I think I was able to distinguish the 2 files with it. The differences are very small. It could also be a fluctuation in the processing chain for digitization. I only said that the ABX test would have more evidence if, for example, an additional recording was checked. e.g. A1, A2, B. If the examiner can then say with certainty that A1 and A2 are the same and different from B, the proof has more evidence than if he can distinguish all 3 with certainty.

 

[pma]

I didn't get an answer to my question about your measurements. Why is there a step down in distortion for the one in green at the high frequency end? Did you not use the same measurement bandwidth for both?

I have overlooked your question. Same BW on both plots, 96kHz sampling, BW48kHz.

@restorer-john asked @pma Did you answer the bandwidth question on your THD vs Freq plot?
Thanks, I have overlooked it.

Step down is because the THD has H3 dominating. H3 measured with BW48kHz is 16kHz maximum.

 

[Sokel]

I doubt you got the point. The ABX test is a precision instrument. Just like with a microscope, the smallest differences can be analyzed. I think I was able to distinguish the 2 files with it. The differences are very small. It could also be a fluctuation in the processing chain for digitization. I only said that the ABX test would have more evidence if, for example, an additional recording was checked. e.g. A1, A2, B. If the examiner can then say with certainty that A1 and A2 are the same and different from B, the proof has more evidence than if he can distinguish all 3 with certainty.

There's in fact 4 choices (files, both real and virtual) to it, A, B, X and Y.
Do the math now.

 

[DaneelOlivaw]

There's in fact 4 choices (files, both real and virtual) to it, A, B, X and Y.
Do the math now.

... and for the test I proposed (which also reveals errors in the digitization chain) you would need A, B, C - X ,Y , Z. Or you would have to do the test three times.

 

[DanielT]

Find some really good jazz recordings made on Soundstream digital gear in the early 80s with the 50kHz SR where they (I think) used the early Studer SRC for 44.1 CD, ones with some cymbals, bell-trees or chime-bar instruments well recorded. HF distortion/harmonic truncation(?) in my experience manifests as a "mush" instead of detail.

It's always a giveaway with compressed audio.

Ask @amirm His jangling keys are a classic and I'm sure they/he had some particular tracks that would show/hide good/poor codecs. Same could be used for this test.

@pma Did you answer the bandwidth question on your THD vs Freq plot?

Here maybe this would be something to test with? In this case in relation to testing what an amplifier can handle in the higher frequencies:

I will use the track Michael Ruff - Speaking in Melodies - Wishing Well (well-recorded fusion) as an example, as there are several high-frequency peaks that require a lot of power from the amplifier. First, the voltage, current and power for the entire signal are shown:

As you can see, we get an active power of 710 Wp at most.

Then we only look at frequencies above 7 kHz (HP filtering with 3rd order Butterworth):

For frequencies above 7 kHz, we get an active power of about 425 Wp. In addition, the strongest high-frequency peak does not occur at the same time as the strongest peak for the total signal.

We can also look only at frequencies above 10 kHz (HP filtering with 3rd order Butterworth):

For frequencies above 10 kHz we get an active power of about 225 Wp. Furthermore, the strongest high-frequency peak does not occur at the same time as the strongest peak for the total signal.

faktiskt.io • Visa tråd - Stereophile recension av Buckeye Purifi Eigentakt

 

[thewas]

Seriously, it was a popular debut album with some catchy hits and reasonably well recorded. It's nothing special otherwise and certainly not "the most revealing for tonal problems". Did anyone seriously say that?

People who have done serious tests, not our anecdotal mood/taste related ones (I also don't like it or am using it ).

A Method For Training Listeners and Selecting Program For Listening Tests

The benefits of training listeners for subjective evaluation of reproduced sound are well documented [1]-[3]. Not only do trained listeners ...

seanolive.blogspot.com

 

[Old_School_Brad]

Here maybe this would be something to test with? In this case in relation to testing what an amplifier can handle in the higher frequencies:

I will use the track Michael Ruff - Speaking in Melodies - Wishing Well (well-recorded fusion) as an example, as there are several high-frequency peaks that require a lot of power from the amplifier. First, the voltage, current and power for the entire signal are shown:
View attachment 443557
As you can see, we get an active power of 710 Wp at most.

Then we only look at frequencies above 7 kHz (HP filtering with 3rd order Butterworth):
View attachment 443558
For frequencies above 7 kHz, we get an active power of about 425 Wp. In addition, the strongest high-frequency peak does not occur at the same time as the strongest peak for the total signal.

We can also look only at frequencies above 10 kHz (HP filtering with 3rd order Butterworth):
View attachment 443559
For frequencies above 10 kHz we get an active power of about 225 Wp. Furthermore, the strongest high-frequency peak does not occur at the same time as the strongest peak for the total signal.

faktiskt.io • Visa tråd - Stereophile recension av Buckeye Purifi Eigentakt

Hmm, that’s an interesting way to illustrate transients requiring dynamic power.

 

[DanielT]

Hmm, that’s an interesting way to illustrate transients requiring dynamic power.

View attachment 443562

But with PMA's op amp test, it could actually be counterproductive to choose test music with a lot of dynamics. Then, if conditions are such, it could be that it is
the potential of the amplifier/headphones/speakers that is tested more than testing whether it is possible to hear differences between different op amps (or am I thinking the wrong way now)?
We also have to hope that those who test are not so stupid that they max out the volume/SPL. Pushing the pedal to the metal and seeing what the amp and speakers can handle can be done another time.

However, PMA's thoughts about using music with a lot of informative, energy in the higher frequency registers in op amp listening test comparisons sound reasonable.
_____
It is good to have a fair amount of amp power headroom because that really dynamic music can be played from time to time. But that is a topic for another thread.

 

[solderdude]

^^^ ooo you actually can tell the difference!

One can only be sure if the test is repeated at least 2 more times by the same person.
One test of 8 is not conclusive.

In fact I would say 20 attempts would be needed and in this case I would spread that over 5x 4 attempts in a row with some time in between.
Listening to virtually nonexistent or very small differences is mind numbing.

 

[pma]

However, PMA's thoughts about using music with a lot of informative, energy in the higher frequency registers in op amp listening test comparisons sound reasonable.

HF content and transients specifically. Sibilants are very good as testing samples. Non-stationary. So not very good for big FFT analysis, which looses resolution then.

BTW, I got 11/16 and 10/16 from the ABX. That makes 21/32 total, you know what it means. Maybe accidentally, maybe not, but DBT ABX is always requested here, for good reasons. Do not expect 32/32 from tiniest differences like this. This is done for fun, but may say something.

 

[Sokel]

HF content and transients specifically. Sibilants are very good as testing samples. Non-stationary. So not very good for big FFT analysis, which looses resolution then.

View attachment 443576

BTW, I got 11/16 and 10/16 from the ABX. That makes 21/32 total, you know what it means. Maybe accidentally, maybe not, but DBT ABX is always requested here, for good reasons. Do not expect 32/32 from tiniest differences like this. This is done for fun, but may say something.

Agreed but I have to point out that, if sibilants are too much as they are at this song, the ones like me who can not tolerate them reject them all together.
Maybe a little less than that, ABX is already boring, it doesn't have to be a torture too.

(God forbid, no Fast Car and stuff like that too! )
((would it be valid if I De-Esser'd the hell out of it? ))

 

[Old_School_Brad]

It could be worse -could be Fast Car with harsh sibilants. Probably listed in the CIA handbook under enhanced interrogation techniques..

 

[DanielT]

Agreed but I have to point out that, if sibilants are too much as they are at this song, the ones like me who can not tolerate them reject them all together.
Maybe a little less than that, ABX is already boring, it doesn't have to be a torture too.

(God forbid, no Fast Car and stuff like that too! )
((would it be valid if I De-Esser'd the hell out of it? ))

Just bite the bullet. If PMA takes the time to fix it so we can test (thanks PMA), then the least we can do is get involved and concentrate on listening. Besides, no one is forcing us to do it. We do it of our own free will because we find it fun and interesting.

After testing is done, we listen to whatever music we want.

 

[Sokel]

Just bite the bullet. If PMA takes the time to fix it so we can test (thanks PMA), then the least we can do is get involved and concentrate on listening. Besides, no one is forcing us to do it. We do it of our own free will because we find it fun and interesting.

After testing is done, we listen to whatever music we want.

Total respect for PMA, that goes without saying.
But sometimes is physical, call it a condition if you like, I have since my teens, so much I have become a human RTA up there

 

[DanielT]

Total respect for PMA, that goes without saying.
But sometimes is physical, call it a condition if you like, I have since my teens, so much I have become a human RTA up there

That was sad to hear.

If we talk about the ability to hear higher frequencies. It's probably mostly old hifi farts, (I'm one of those) here at ASR who either due to natural age reasons don't hear that high up frequency anymore. Or who have had hearing problems, impairment because hearing protection wasn't used to the same extent in the past as it is now.
However, I suspect that when PMA mentions HF content, he is referring to up to, probably around 6kHz-10kHz, or @pma ? These are the frequency levels most of us can hear.

Maybe there are a lot of younger people reading this thread and testing, who knows. Those of you who are twenty-five years old or younger make yourselves heard in the thread.

 

[w277]

Unfortunately 8 trials are not sufficient, 10 is the recommended minimum.

ABX test - Wikipedia

en.wikipedia.org

9/10 is considered "statistically significant" but is still not absolute proof: there is still a residual (around) 5% chance that it was pure guessing.

If you want something close to proof of audibility, you need to be targeting a p value of 0.01 (1%), or even lower.

In my modest understanding, statistics is here not about "absolute proof".

If you have 7/8, the p value is 3.52% (as stated in the abx report above). 9/10 is 1.1%. // 10/10 is 0.1%.
For many purposes, a p value of 5% is considered acceptable.

The value of ten trials comes (according to your reference) from a manufacturer of an analog abx comparator, who seemingly also suggested not doing more than 25 trials.
For me with that source material (no blame on PMA!), ten trials would be painful.

In order to gain a broader view, I would like to invite other members to provide their (detailed) test results (irrespective of the number of trials and result).

 

[antcollinet]

Lastime I looked the ABX addon for mac was not available. Now it is.

I thought I could hear a difference at the start of the applause. Looks like I was wrong. I'll try to find differences elsewhere.
(HD560 headphones, Headphone out from Denon AVC3800)

foo_abx 2.2.3 report
foobar2000 v2.24.3
0125-04-11 17:04:25

File A: 2134.wav
SHA1: be3c81c4a4dd185262c511bcb0447d7f8c6da138
File B: 49720.wav
SHA1: ec38050c91e3d36050a8a4d580c51324ca3beb4f

Output:
Core Audio : Default, 16-bit
Crossfading: NO

17:04:25 : Test started.
17:05:38 : Test restarted.
17:05:38 : 01/01
17:06:34 : Test restarted.
17:06:34 : 01/02
17:07:07 : Test restarted.
17:07:07 : 01/03
17:07:43 : Test restarted.
17:07:43 : 02/04
17:08:48 : Test restarted.
17:08:48 : 02/05
17:08:59 : Test restarted.
17:08:59 : 03/06
17:09:12 : Test restarted.
17:09:12 : 04/07
17:09:25 : Test restarted.
17:09:25 : 04/08
17:09:33 : Test restarted.
17:09:33 : 04/09
17:09:57 : Test restarted.
17:09:57 : 04/10
17:09:57 : Test finished.

----------
Total: 4/10
p-value: 0.8281 (82.81%)

-- signature --
9a8ac22c65cf4b032aebd7224ae39524e58559d7