Hello,

I have problem which probably can't be solved, but I'll ask anyway. There are people claiming they can always differentiate a 320 kbps MP3 from original CD. I don't think they can and I'd like to prove them wrong. What's worrying me is the possibility of spectral analysis which can always tell, which one is the original and which one is the compressed sample (even after conversion to a lossless format).

Is there any solution? I cannot imagine one, but maybe somebody will have an idea...

You'll hate me for suggesting it, but encode both as DRM protected WMA lossless samples. Makes it much more difficult to get at the decoded bitstream for analysis, but you can still listen.

I'm not so sure how a spectral analysis would always be able to tell that.

It would be better (and simpler) to do a double blind ABX test, like with foobar's ABX component or ABC/HR. This would require some software on these user's end, so it won't probably be the most practical solution... but protecting audio against FFT seems impossible if not, as posted, by <spit> DRM.

It sounds as if vlada is giving the files to someone over the internet (an audiophool) to do the test themselves and wants to avoid any "cheating" by looking at the spectrum and unblinding the test.

But if you send to someone WMA protected files, how could he simply listen these files? Aren't they, well... protected

He could still run Spectrum Lab (or similar) in the background, you'd need a version of ABC/HR that uses encrypted samples and constantly scans all running processes for 'cheating tools' (just as afaik some games do).

At 320kbps you could simply disable the lowpass.

It is still quite easy to distinguish 320 kbit mp3 from lossless based on a spectral view even with the lowpass disabled.

Distinguish the two, sure. Identify which is which, not so sure. Convince me.

At least not with Lame (the psymodel is limited to 20k)

There should have been good, opensource tools for distributed blind-testing.

A simple Java-application that demanded exactly no user intervention, packaged with a number of audiofiles that are simply copied to a local folder and program started, autogenerating a mail for every run,

Php-script or similar for doing the same thing on a webpage.

I HAVE BEEN LOOKingfo that kind of stuff.

-k

No sweat Canar. Give me two files, I'll tell you which is which. You guys act like this is difficult.

It isn't.

I'm afraid that even protecting the files with some DRM or encryption won't help. Why? The user can always digitally record the output of a sound card and analyze it. I already wrote in the beginning that there is probably no solution for my problem. But somebody might come with an idea I haven't thought of.

Forgive me if this is off-topic, but this thread brought it to mind---

Recently an "audiophile" friend of mine insisted he could always spot MP3 encodings since they "destroy the fidelity".

I encoded a selection of some classical flac archives I have with lame to CBR 160. Then, I decoded them back to wav, and converted back to FLAC. Then I transcoded the original FLACs once more, but this time to -V 0 --vbr-new.

For only god knows what reason, he was telling me how much fidelity and "sense of reality" were lost in the lame -V 0 encoding, but swore that all those problems "instantly vanished" when he auditioned "the flac source" (which was actually CBR 160 !!!)

Thought you all might enjoy that

(Sidenote: I was able to ABX all of the samples I sent him at 160cbr vs -V 0, the later of course being transparent)

Modern soundcards have a way to determine what is connected to an output. Maybe it is possible to create application which would send audio to only one jack (output) and make sure that headphones are connected to that jack.

Are these people worth such an effort?

Just try SirChristof's Method to proof them wrong

If one has complete control over enviroment then test that SirChristof proposes can be done. But if that is not possible then tester might find out by spectral analyses which settings for an encoder were used.

What you're trying to do is exactly what every music distributor is trying to do as well... create an unbreakable DRM scheme, and you all know it can't be done.

If people want to live inside their own little ignorant bubble, and make ridiculous and unfounded claims about perceptual coding, just let them. You can't save everyone, and you certainly can't save someone who doesn't want to be saved. If they were genuinely interested in testing their own claims vs. reality, they wouldn't even try to cheat on the test.

Supose that a company wants to win in some public listening test.
It has the money and capabilites to get many PC's which would serve as the spectral analyses tools.
How to stop it?