While skimming through a new re-release of a Cher album of love songs from 1971, I came across one particular half-minute segment with some extreme high frequency audio content. I actually forgot which song it's from, as I ripped the audio and then gave the CD back to the person I borrowed it from, but here is an unmodified, uncompressed WAV file I made of that particular segment (4932 kB):

http://rvcc2.raritanval.edu/ktek9053/cherclip.wav

I do not suggest listening to this sample repeatedly at high volume levels!! It contains a piercingly high whooshing noise the sweeps back and forth from side to side, combined with very bright strings and other treble-intensive sounds. And as you would expect from a modern "Digital Remastering", it is highly compressed and equalized to emphasize this bright, strident sound. The rest of the song is actually quite "normal" -- just this half-second intro is very unique!

In fact, its spectral graph (of the Left and Right channels of the original WAVE file) almost looks like artwork! (225k GIF file):
http://rvcc2.raritanval.edu/ktek9053/wav.gif

As you might expect, this is a very challenging sample for "lossy" encoders to deal with! For example, here's the result of encoding it with Fraunhofer's "Fast" MP3 codec, at 128 kbps (187k GIF file):
http://rvcc2.raritanval.edu/ktek9053/fast128.gif

Yikes!! I have found that Fraunhofer's original "L3ENC" codec -- dating back to 1996 -- actually provides the best results at 128 kbps, even compared to LAME 3.90 at "-q 0"! But still, *anything* at MP3-128kbps sounds very bad with this sample.

So, fool around with your favorite encoders, with your favorite "tweaks"... just don't listen too much, or it'll hurt your ears! That's why I suggest using Spectrum Analysis (like I did in Cool Edit Pro), so you can *see* the results even without hearing them.

And feel free distribute this file to other appropriate message boards and web sites... just please upload it to your own servers, so mine won't be overloaded with all the downloading.

As you've been a member for > 1/2 year you probably know most of the things I'll write but your post makes me doubt you do, so I'll tell anyway:

- Statements about sound quality should be backed up with double blind (= ABX) tests (and - in your case, using an outdated, hard to find encoder - providing a sample compressed with this one would be a good idea). This is a forum rule.

- Spectrum analysis might be useful additionally, but it can't replace listening tests. You can't see how it sounds.

- lame 3.90 128kbps with -q 0? Why don't you use --alt-preset cbr 128 to get best quality out of lame at 128kbps CBR? You might want to have a look at the sticky "list of recommended lame settings" thread. If you have used this setting and you hear obvious problems, please describe them, tell at what position they occured and provide your ABX results.

If you don't know what I'm talking about in some points I suggest reading the FAQ or using the search function.

You're right about high frequencies, especially loud high frequency noise being hard to encode.