Does what we hear correspond to what we measure?

[Sir Sanders Zingmore]

Do you understand that differences in volume do not manifest themselves ONLY as a loudness difference when it comes to human hearing perception? Can you understand how a volume difference could therefore lead to misleading perceptions of differences in sound quality?

This ^^

@Vintage02 it’s almost certain that when levels are not matched, people will report differences that completely disappear once levels are matched (for example, “dynamics” or “clarity”)

 

[Timcognito]

These kind of posts are frivolous placebo exercises. But your time is just that, your time.

 

[Sono]

Because based on a quick listen to part of each track, neither one sounds bad to me - I hear no unpleasant distortion, and there's either no significant noise, or if there is noise it's masked by the background noise of the venue where it was recorded. The spatial mix sounds essentially the same on both, none of the instruments or vocals sound grating or unpleasant in either version, and the reduced dynamics on #1 are in my perception modest and not remotely egregious. EDIT: I didn't find them to sound quite as similar as @levimax did, even when level matched. But the spirit of levimax's comment is something I agree with - just because we can hear a difference doesn't mean we inevitably have a preference - it doesn't inevitably mean that the difference we can detect is significant to us.

If I'm listening at low to moderate volume, I might prefer #1 for its added bass impact. If I'm listening at higher volume or hypothetically listening to the entire live album that this song comes from, I might prefer #2 because it might be less fatiguing over time and the reduced dynamics of #1 might get annoying or unsatisfying over time.

So who cares which one I prefer - and honestly, who cares which one you prefer? I've given @Vintage02 and anyone else who's interested plenty of information about what sonic differences I hear. Beyond that, which one of those two sonic profiles I might prefer is a meaningless question. What matters to me is, do they sound different and if so, how or in what ways do they sound different.

So we have at least two »Major Contributors« with us who are not able (willing?) to chose a preference here?

 

[Sono]

This is a title by the French artist Michel Jonasz : 'Les fourmis rouges' taken from a 2017 live performance.

Why is there a single hi-hat opening sound at the beginning of #2 that is missing at the beginning of #1?

 

[levimax]

So we have at least two »Major Contributors« with us who are not able (willing?) to chose a preference here?

For me at least I need a difference to form a preference and both sounded fine. It's not like some remasters with huge compression or EQ changes. If this was a track (or album) I really enjoyed and listened to a lot and cared about I would spend some time listening to both versions and probably end up preferring one but nothing in the short time I listened jumped out enough for me to form a preference.

 

[levimax]

Why is there a single hi-hat opening sound at the beginning of #2 that is missing at the beginning of #1?

#2 is 2 seconds longer and starts a little earlier than #1. Now you have the key for a very easy 100% success in a level matched blind test

 

[Sono]

Would be quite baffling to know if this single hi-hat opening is really enough to make people (like me) preferring that version ...

 

[tmtomh]

So we have at least two »Major Contributors« with us who are not able (willing?) to chose a preference here?

This comment would make some sense if I (a) had said I couldn't hear any differences between the two or (b) had failed to explain the reasons I don't have a clear preference and the scenarios in which I would prefer one or the other. But I explained that I do hear differences, and I explained what I perceive those differences to be, and I explained precisely why I don't have a clear preference, and what I suspect might make me prefer one or the other depending on the listening scenario.

You're clearly trying to insinuate something - so how about just say what you want to say?

 

[Sono]

This comment [...]

… was a simple yes-/no-question. Shouldn’t be a problem to answer for anybody involved.

 

[tmtomh]

… was a simple yes-/no-question. Shouldn’t be a problem to answer for anybody involved.

Why does it upset you so much that folks can detect audible differences but those differences don't necessarily add up to a clear preference?

 

[krabapple]

Hello, mystery person. You ask:

Does what we hear correspond to what we measure?

Is this really what you are trying to address?

You've already been informed of the common psychoacoustic case where hearing something as 'better' corresponds to it measuring a bit louder (the result of simply 'turning it up' , or applying compression)

So? Is the answer 'yes' or 'no' in that case? You tell us first.

 

[levimax]

Would be quite baffling to know if this single hi-hat opening is really enough to make people (like me) preferring that version ...

I wasn't talking about preference, I was talking about ABX tests. In the objective world you have to "prove" you can hear a difference before you can claim a preference. The way to do that is with a level matched, blind listening test. The objective crowd likes to see at least 80% out of 18 tries and 100% is better. The easiest way to take these tests is with software that level matches the sound and then you can listen to sample A and sample B as much as you like and then the software presents sample X and all you have to do is say if X is A or X is B. You have to do it 18 times and you don't know your score as you go. It can be much harder than you think and often fatigue sets in.

My comment was that if you figure out a way to "cheat" i.e. you know sample B has a high hat at the start and sample A doesn't then all you have to do is listen to the start of the track X and you will know which sample it is. This is easy and you can score 100 % without much effort. The classic example of this is people that claim to hear a difference between 44.1 Khz 16 bit and 192 Khz 24 bit. Most likely it is not possible but what is possible in some cases is if you can find a quiet part of a sample and turn up the volume super loud the 44.1 Khz sample with have more noise (hiss) than the 192 Khz sample. You need to be careful because if you aren't quick with the volume you can blow your speakers or if using headphones it can be painful and dangerous to your hearing.

 

[ahofer]

I don think measurements are like wine labels. They would be more like using chronatography to see the wine chemistry and compare it to a reference composition.

The labels are like….the badges on the components, or their brochures, I.e……the labels.

Studios introduce all sorts of distortions and EQ to achieve a sound. That’s the art. Audio measurements tell us what other distortion and EQ the reproduction system might add.

This conversation always seems to veer into “audio reproduction as art form” territory, which I find bizarre.

 

[Mneff]

Hello,

You and I are fans of the magnificent reviews done by Amir, NTTY, and others, where they show us measurements that seem to predetermine what we are going to hear.

But what if we did the opposite : let's listen and try to correlate what we hear with objective measurements.

First of all, I suggest that you listen to two recordings of the same piece and already share your opinions on which one seems to be the 'best'.

This is a title by the French artist Michel Jonasz : 'Les fourmis rouges' taken from a 2017 live performance.

Obviously, some will be tempted to compare the files using programs like Audacity or DeltaWare, but I would like you to just listen on your system or through headphones and tell me your preference for No. 1 or No. 2

Here are the links for these pieces :

N° 1 : Michel Jonasz - Les fourmis rouges (Live Casino de Paris 2017)

N° 2 : Michel Jonasz - Les fourmis rouges (Live Casino de Paris 2017)

Then I will give you more details about the two recordings.

Thank you in advance for your participation

2 kinds of live music. Amplified and unamplified. The latter is almost always opera or orchestral or classical. The sound that a violin bow makes as it is dragged over the strings goes directly from that point to your ears in a live setting. Trying to record and recreate that sound so that the recording sounds “live”is very challenging. Once amplification is injected in to the chain, all hell breaks loose.

 

[tmtomh]

Why is there a single hi-hat opening sound at the beginning of #2 that is missing at the beginning of #1?

Because he wasn't careful or precise in creating the two different samples. But remember, he doesn't want to explain anything to anyone, he just wants people to say which they prefer - and you have been quite insistent about that too, so stop asking about the hi-hat and GIVE YOUR PREFERENCE OR ELSE!!

 

[RexrothPigeon]

I don't want to listen to that music, just not what I want to put in my ears

 

[Hayabusa]

to correlate measurements with what we hear we need to add psycho acoustics to mix.

 

[MAB]

To determine preference, first a listener needs to be able to establish a difference. Preference tests are massive studies. Quick checks like this are not useful, they even sometimes lead to results counter to actual preference. This type of outcome isn't unique to audio.

I have now hundreds of folders of WAV files posted by people on the internet attempting to do science Dunning–Kruger style. Some of the files are clearly manipulated, in obvious bad faith to try to generate some sort of gotcha moment. The majority of them are just odd. Missing the intro in one track is odd, not sure if this was on purpose or if care had not been taken. The level matching is a real problem.

Regarding difference, since we need to walk before we run... I trimmed off the first seconds of track #2 to eliminate that tell so not to as not to fake the result. In the process of trimming, I notice that track #2 #1 has 880 ms of white space inserted where the hi-hat decay should be, which actually makes me suspicious that the files are actually manipulated. Putting that aside, without the tell of the missing hi-hat the differences in levels is sufficient to tell the difference 20/20.

Here is the test result validation:

Spoiler: ABX result, no level matching, trimmed intro

foo_abx 2.2.1 report
foobar2000 v2.1.6
2025-10-18 20:28:26

File A: 01-Michel Jonasz - Les fourmis rouges (live Casino de Paris 2017).wav
SHA1: 6bc70cc4198cf8c521b081cb7cae7e982c4b721f
File B: 02-Michel Jonasz - Les fourmis rouges (live Casino de Paris 2017).wav
SHA1: 0e32eb6f345761a0dd052ac5e65159b04ef3d5dc

Output:
Default : Primary Sound Driver
Crossfading: NO

20:28:26 : Test started.
20:28:49 : Test restarted.
20:28:49 : 01/01
20:29:04 : Test restarted.
20:29:04 : 02/02
20:29:13 : Test restarted.
20:29:13 : 03/03
20:29:19 : Test restarted.
20:29:19 : 04/04
20:29:25 : Test restarted.
20:29:25 : 05/05
20:29:31 : Test restarted.
20:29:31 : 06/06
20:29:36 : Test restarted.
20:29:36 : 07/07
20:29:42 : Test restarted.
20:29:42 : 08/08
20:29:52 : Test restarted.
20:29:52 : 09/09
20:30:06 : Test restarted.
20:30:06 : 10/10
20:30:11 : Test restarted.
20:30:11 : 11/11
20:30:18 : Test restarted.
20:30:18 : 12/12
20:30:23 : Test restarted.
20:30:23 : 13/13
20:30:38 : Test restarted.
20:30:38 : 14/14
20:30:46 : Test restarted.
20:30:46 : 15/15
20:30:52 : Test restarted.
20:30:52 : 16/16
20:30:57 : Test restarted.
20:30:57 : 17/17
20:31:02 : Test restarted.
20:31:02 : 18/18
20:31:09 : Test restarted.
20:31:09 : 19/19
20:31:16 : Test restarted.
20:31:16 : 20/20
20:31:16 : Test finished.

----------
Total: 20/20
p-value: 0 (0%)

-- signature --
2b42873861451d0ffc058e0ae56ca1953957f323

This result is with desktop speakers, and was one of the easier ones in recent memory. The level difference is quite significant, and in no way can I judge what else is going on. The need to level-match is literally covered in in chapter 1 of Floyd Toole's book. Might be good to get the fourth edition which just started shipping.

I also notice that track #1 is inverted relative to track #2. This doesn't affect the audibility.

Back to preference. First, I need to judge difference. I level matched the files. I tried four trials. First with speakers where I scored 9/20, second trial was with headphones, and I scored 13/20, and I thought I was onto something, but the final two trials with headphones were 9/20 and 11/20, so not able to detect difference when level-matched.

I note that few who actually pass an ABX test never posts their results. @pma recently posted a poll using some expertly generated files for comparison, there were 11 positive response to the poll yet nobody posted their ABX test report as requested. The part that I can't tell in the case of this thread is why the files are edited the way they are. There are differences beyond just the missing hi-hat.

edit: got the track with the white space wrong

 

[Vintage02]

Because he wasn't careful or precise in creating the two different samples. But remember, he doesn't want to explain anything to anyone, he just wants people to say which they prefer - and you have been quite insistent about that too, so stop asking about the hi-hat and GIVE YOUR PREFERENCE OR ELSE!!

That's exactly it. I do not claim to know or to teach anyone, I am questioning the relationship between the technical measurements that can be made and what they might imply during listening, or at least up to what level. I know that playing a few dB louder makes the perception 'better.' Regarding the two recordings, indeed there may be a slight offset, and you will understand when I specify which device it is.

 

[Vintage02]

I also notice that track #1 is inverted relative to track #2. This doesn't affect the audibility.

Back to preference. First, I need to judge difference. I level matched the files. I tried four trials. First with speakers where I scored 9/20, second trial was with headphones, and I scored 13/20, and I thought I was onto something, but the final two trials with headphones were 9/20 and 11/20, so not able to detect difference when level-matched.

I note that few who actually pass an ABX test never posts their results. @pma recently posted a poll using some expertly generated files for comparison, there were 11 positive response to the poll yet nobody posted their ABX test report as requested. The part that I can't tell in the case of this thread is why the files are edited the way they are. There are differences beyond just the missing hi-hat.

The recordings are as they were recorded; I just tried to align the two tracks, but it's possible that I missed the mark there. I could have taken more care with this or adjusted the levels to make them match, but I wanted to do as little editing as possible to stay as close as possible to a listening experience that I would call "real." That is, as if someone were playing them for me live without any manipulation.