A lot of questions and discussions are about transparency or which settings can be considered as generally transparent.

"Do an ABX-Test for yourself" is a common answer. But this isn't quite fair, as the hearing can be trained and first results may turn out as inaccurate later.

So I am thinking if it makes sense, to do some public listening tests that determine both, transparent settings for lossy codecs and when they start to sound annoying for most listeners.

A very high (or the highest) setting as high anchor, a very low setting as low anchor an 3 mid-range settings, that depend on experiences with a codec.

For Lame-MP3 this could be V1 (high anchor), V4, V5, V6 and V9 (low anchor).

The testsamples should contain known problem-samples.

What do you think?

The problem is, transparency -- by its very definition -- is different for each and everyone of us. What's transparent to one may easily be non-transparent to others.

It's like determining "The Best Tasting Ice Cream in The World" -- Vanilla? Chocolate? Strawberry? Green peas?

(Ice cream analogy borrowed from a recent thread in HA)

The fact that everyone is different won't prevent you from determing an average, and getting settings where you can say "this is transparent for 80% of the people out there...95%....99%...."

Ha, that brings us to the second problem: 80% of what? Of the 6 billion people on the planet? Of HA members? Of HA members willing to participate? Can we extrapolate the last one to cover the first one?

And, echoing elmar3d's concern -- that number over time will get smaller, for the exact reason that he/she posted: People (without proper de-learning) will learn about artifacts, raising their sensitivity.

Which is why I give up doing transparency test. I just encode/transcode, listen, if the result seems to have an artifact, play the original at the same location (NOT ABX). If the original has the same seemingly-artifact, return to the result file. If the file is acceptable... to hell with transparency

BUT... I still challenge people to ABX if they start making strange claims such as "... any lossless compression will reduce soundstage making the tracks seems less warm, lose color, blah blah blah" (*cough* audiophooles *cough*) I love it when biased people claiming golden ears have to eat their word (foot included).

I also think that such a test would have to be carried out for each major codec. Also, you would need CBR vs. VBR tests to decide which is better.

Result of such test will also depend on the kind of tested samples.
"Transparency" has at least two different meanings:

- transparent for normal usage (i.e. when people are normally listen to music)
- transparent in a more critical or absolute way.

On one hand a setting could logically be said as non-transparent as long as one sample could be ABXed by the tester. I can remember some "guru" --especially on trading networks -- who based their "absolute" (and of course "reference") settings on the existence of one killer sample (something similar happend to me, and I really started to think lossless once I ABXed 2 or 3 samples with MPC -q10).
On the other hand and on a practical point of view it rarely makes sense to focuse so much on rare exceptions -- and a given setting could be said as transparent even if the subject knows some samples which aren't.

Both approach could nonetheless make sense for a listening test. The first one would satisfy people looking for "near-lossless" quality at friendly bitrate (i.e. most people using ~192 kbps and more aka paranoiac persons); the second one is more for reasonable people looking for an immediate satisfaction.

I doubt people who are using 192 kbps are going to switch to lower bitrates anyways. I saw people on various forums claim that at 128 kbps, frequencies that are perceptible are cut. Pointing them to my 128 kbps test didn't help and they always had something to complain.

Hopefully this isn't offtopic, I think it's worth mentioning.

I find that, in the vast majority of cases where the (mp3)bitrate is, say, anything above 128kbps average, 99% of all "defects", "artifacts" and "nasty unwanted things" present in the file----were a result of something other than the lossy compression scheme. Tape hiss, line noise, bad mixing, crappy mastering, bad mic placements, the list goes on. And somewhere, way way down the list at the very bottom of "things that can ruin the sound", would be "lossy compression".

There have actually been a number of times when, listening to an MP3, I heard some type of artifact or "problem". I would get the original CD or FLAC archive, and compare it to the original. Sure enough, that same "artifact" was present in the original. Go figure.

I would venture to say that, if you are listening to music with the goal of hearing artifacts & problems, you will find many of them-------even if you encode everything losslessly.

So do keep that in mind when thinking about all of this.

Regarding the problem of defining "Transparency": My intention was to do this after the testing.
This could be done by determining thresholds for the variants of "Transparency" that guruboolez mentioned, for example.

At least as far as HA is concerned, transparency has a working definition: http://wiki.hydrogenaudio.org/index.php?title=Transparency
and I don't see any problem with it.
As for listening tests, by definition ABX tests cannot prove transparency for codec X at a given bitrate Y, they can only demonstrate no-transparency.
Reading the definition of "Transparency" linked above would have avoided this thread entirely, or at least made it much shorter...

I really love this "snap up a word, link to a wiki , stop the discussion"-behavior!

Actually it was more like: check the excellent HA Knowledgebase first.
I didn't think this simple suggestion would make you feel so unconfortable...

By definition, it doesn't.
It's also, I think, what Pepoluan was trying to tell you.