A friend of mine (who cannot post here because he only has yahoo and gmail emails) want me to post this:

"Transparent" means that you can't tell that the sound you're hearing has been compressed for storage and subsequently decompressed for playback. It's analogous to looking through a window. If you can tell that the window glass is there, then the glass is not transparent. If you think you're looking straight out into the outside world, then the glass is transparent.

A simpler way of saying this is that "transparent" means your ears can't tell the sound was compressed.

If you can't hear any difference between -preset medium and your original lossless recording, then -preset medium is transparent for your ears on your equipment for the type of sound you're trying to compress, and you should therefore keep using it.

I'm sure he understands what transparent means, but rather he (and I too) want to know exactly what is meant by "generally transparent" as used in the List of recommended LAME settings FAQ (at the top of this forum). It could mean anywhere from 70% to 99.9999999%. With all the stress here put on providing numbers to back up statements about quality, it's somewhat ironic the FAQ doesn't supply any data.

Given that its not possible to test all possible audio files with all possible LAME settings and all possible listeners, saying that its "generally transparent" is as good as can be done.

So you're saying since it's impossible to test all music on everyone, then no foundation has to be provided? That doesn't seem to jive with TOS #8.

The foundation is in the listening tests that were performed several years ago when the alt-presets were being developed. You could search those threads and find out what threshhold they were looking for.

Nice straw man argument by the way.

I understand what your getting at and I'm sure there could be some 'number' derived as
there are threads listing samples tested and the outcome.

I just don't think they are easily totaled or calculated anywhere.

Of course 'Generally' also does mean not specific

P.S. By the way what's up with?

You cannot create an account here using a yahoo or gmail account.

Apologize for taking this thread off topic..



Did not know that... Why is that?

I suspect it's to keep people from creating bogus/multiple accounts here (and they feel it's too easy to setup a bogus yahoo/gmail account).

Although I don't have hard figures to back up my claims, and although this isn't something that's generally mentioned by people at HA, I've heard people say that...

(dramatic pause)

Your equipment shouldn't be much of a problem, because enhancing your equipment doesn't enhance your hearing. When it comes to lossy audio, what matters more is previous experience, training, and familiarity with compression artifacts.

I like to make analogies between lossy compression and magic tricks. You can't figure out what the magician is really doing by getting a bigger TV.

Hmmmm - I find that I can't distinguish --preset standard from extreme on my speakers (Harbeth BBC monitors driven by tweaked Quad 606) but with my iRiver H340 and Etymotic ER4-S phones, I can tell 12 times out of 15, with a preference for extreme.

Music
Pink Floyd Meddle: Fearless
Led Zeppelin Remastered: Whole Lotta Love
Zappa One Size Fits All: Sofa No 1.

Testing setup:
No computer ABX, but friend behind me out of sight, playing first 20 seconds of each track, with first selection determined by toss of coin. On the iRiver this was achieved by creating playlists of two tracks, one aps the other ape, both ways round, and selecting the second track after 20 seconds. I have one guess based on whether I think the articulation and presence has improved or got worse. If I think it has improved I guess extreme.

This produced the following results with the Etys:
Pink Floyd: 3 correct guesses out of 5
Led Zep: 5 correct (!)
Zappa: 4 correct

I haven't tested for significance but it seems likely that this would be highly significant. With speakers, my guesses were no better than chance and subjectively I felt I couldn't tell the difference.

This is STL's friend. I found an old e-mail address I haven't used in 5 years that the system would take for verification.

I think my question might be misunderstood. I'm not asking for somebody to tell me if a particular LAME preset is transparent to me. I can test that myself in an objective manner. I'm also not asking what the threshold of transparent is, rather I'm just curious what the words "generally" and "most" mean in the statement "should generally be transparent to most people on most music" from the LAME Preset Documentation.

I believe I understand what the statement is supposed to mean to the average user. To me it says "unless you're advanced enough to test the transparency yourself you are unlikely to be disappointed with this preset". I'd like to know what the LAME developers had in mind for that statement from a statistical point of view.

I'm sure the answer to the question is in the forum archives someplace but I've been unable to find it. The question is purely to satisfy curiosity, but can somebody point me in the right direction?

It just means that words are over-carefully choosen to exclude the possibility of people complaining latter about a potential problematic sample that they might find.

In real world, this preset is quite safe although there are always some occasionnal problematic samples.

I don't think there are any solid statistics. But it does seem that it must be at least 90% / 90%, ie transparent for 90% of people on 90% of music. And it's probably greater in fact. Otherwise we would have many more reports of preset standard failures. Out of those that do ABX tests here on HA, there are only a few people that can successfully ABX preset standard with regularity. There are also cases where more people can ABX preset standard, but those can be assumed to be problem samples.

One of the problems with trying to say anything more definite than "most people for most music" is the lack of actual numbers. The reason we point to about the lack of numbers is that it is very difficult to conduct high bitrate listening tests. Not enough volunteers will be able to ABX the samples. So paradoxically, the fact that preset standard (and other codecs at high bitrate) is so good means it is difficult to quantify exactly how good.

Personally, I take that as good sign.