Proof that high resolution audio sounds different

[jkeny]

This won't come as a shock to many here but it could be an important moment in this audio hobby - the first "valid" blind test results are coming in that high-res audio sounds different to 16/44.1. See here. The test files are also there for people to try them themselves.

Now why I say "valid" is because it is using Foobar ABX testing which basically is considered valid by all those who clamour for blind tests to "prove" you heard what you say you heard. This Foobar utility works by you selecting two audio files which you can play at any time. One file is 24/96 & the other is the downsampled 16/44.1 version of this file. ABX utility randomly chooses one of these files as X (you don't know it's choice). Your job is to identify X as either the first track (24/96) or second track(16/44). It keeps track of your correct choices & produces a statistical accurate score at the end of the test. For anybody who has studied statistics you will know that the score shows whether your choices were pure guess work (~50% score) or that you could actually hear differences & identified X correctly (0% score).

Of course these results have caused a bit of a stir & other possible reasons for being able to correctly identify the 24/96 file from the 16/44 file. Things like distortions in the playback system that reacts differently to 16/44 Vs 24/96. distortions that might have been introduced in the downsampling to 16/44, cheating & some other reasons. So far none have held up to scrutiny.

But just to be clear, these test results do NOT mean that people can hear >20KHz frequencies. These test results do NOT mean that 24/96 is PREFERRED to 16/44 - preference is not being tested

What it seems to show is that some training of your hearing is required to focus in on specific areas between the two files that reveal the differences.

Hopefully, it's of interest - have a read of that thread. There's another one on AVS forum but it is full of bile.

[Claus]

I have mentioned it before, but I could easily tell 24 bit as better vs 16, when recording in my basic home studio. But sample frequency is much harder to hear. Hence I always record in minimum 24 bit. I Leave the frequency at 48 to save space...

I doubt anyone can hear the difference between 24/96 and 24/192...

[Claus]

If anyone is interested I can record a few simple tracks at different rates to upload for forum users...?

[jkeny]

Hi Claus,
The files used in this test are a recording of keys jangling which is chosen because it has a lot of energy in the frequencies >20KHz to ~40KHz (much more energy in these frequencies than any musical instrument). It was recorded at 24/96 then downsampled to 16/44 (which knocks off any frequencies >22KHz) & upsampled back to 24/96 - this is done so that the DAC uses the same internal pathway when playing back both files. Normally the 24/96 pathway is different to the 16/44 pathway in a DAC with different filters being enabled & this can cause an obvious audible difference between files.

The files are here

The Foobar ABX utility is here

[nige2000]

so how do i know if i got it right?

[jkeny]

so how do i know if i got it right?

When you run the Foobar ABX utility it shows the two files you have chosen to test for audible differences as File A & File B. For each trial, ABX randomly chooses one of these two files to be File X. You job is to correctly identify X as either sounding the same as A or the same as B. You can play A, B or X any number of times, one after another, the same snippets of each one, etc. When you are happy you choose File X as sounding the same as either File A or B.

The ABX utility keeps score of your choices. The idea is to repeat the trails enough times (>10) so that some statistics can be calculated which can tell if you were guessing or actually hearing differences. If guessing your score will come out as around 50%, if you got all choices correct it will give a score of 0% i.e. you are choosing based on hearing actual differences (or you are extremely flukey :))

[jkeny]

Oh, btw, I just came across this Head-fi thread where a challenge to identify differences between DACs was posted. 6 files were posted that represented the recordings of 6 different DAC's playing back the same song. Not a great test, I know - lots of variables & possible issues. But some people posted positive ABX results - proving that they could reliably hear the difference between two recordings (DACs?).

One of the reasons I'm posting these things is so that I (& possibly others) have a marker for links to results that can be offered when anyone asks for blind test proof of differences (between DACs, high-res, etc). Whatever the final reasons for the positive results of these tests turns out to be.

I only came across this Head-fi thread because of Maxflinn post on Whatsbestforum (WBF) asking his usual "So has anyone ever successfully differentiated two (or more) DACs in a blind test before?" A quick search revealed the head-fi thread - so much for people being objective & uncovering evidence, themselves.

I'll repeat here what I posted on WBF
Reading through that Head-Fi thread is another interesting study in psychology - just how we all are prone to accepting evidence that favours our position & rejecting evidence that doesn't (this seems to be a bigger bias than sightedness ). The thread starts off by the o/p in a very bullish way

"Objectivists claim that all DACs that aren't severely broken sound the same.
Audiophiles claim that different DACs sound different, more or less noticeably. I've been told, for instance, that a Clip+ and an iPod Classic sound completely different.
......
You may try to identify a particular piece of gear, or which sample is the original, or rank the samples in descending order of preference (the one you like the most first, the one you like the least, last), or ABX any two samples… whatever you're most comfortable with.

P.S.: I don't expect that many people will venture a guess or two, but at least the opportunity is here.

And ends up after some positive ABX results are posted saying

"Knowing what to listen for gave you an advantage that you wouldn't have had, if you hadn't analyzed the files first. And for the same reason, you didn't choose which files to ABX by accident. Not exactly 100% blind, is it.

Interesting, isn't it, as this is repeated on nearly every forum where "objectivists" post. I even started a thread on PFM called "sorting out evidence-based from faith-based" & even though the thread's objective was in the title people showed their "belief system" by their very posts trying to argue that the ABX results were "faked", "needed proctoring", "needed repeating", "result of IMD", "result of bad downsampling", etc.

Thanks, Max for making me do this search - I wouldn't have read that Head-fi thread otherwise

[nige2000]

so how do i know if i got it right?

When you run the Foobar ABX utility it shows the two files you have chosen to test for audible differences as File A & File B. For each trial, ABX randomly chooses one of these two files to be File X. You job is to correctly identify X as either sounding the same as A or the same as B. You can play A, B or X any number of times, one after another, the same snippets of each one, etc. When you are happy you choose File X as sounding the same as either File A or B.

The ABX utility keeps score of your choices. The idea is to repeat the trails enough times (>10) so that some statistics can be calculated which can tell if you were guessing or actually hearing differences. If guessing your score will come out as around 50%, if you got all choices correct it will give a score of 0% i.e. you are choosing based on hearing actual differences (or you are extremely flukey :))

Sounds time consuming
Played the two tracks on foobar
Sound different on my desktop and "SOUNDBAR" but not as much difference as I thought there'd be

[jkeny]

Sounds time consuming
Played the two tracks on foobar
Sound different on my desktop and "SOUNDBAR" but not as much difference as I thought there'd be

The test itself really shouldn't be time consuming but identifying where in the file is best to hear the differences between the files could be - if that is the way you want to do the test i.e. it's a bit like spot the difference visual tests - takes a while to find it but then it's obvious. The other way I've seen people pass the test is not to do this sort of focussed listening but to intuitively pick. Some people, I believe, surprised themselves that they got it right without knowing what differences they might be hearing (they did it enough times to statistically rule out fluke)