Re: [MLUG - DISCUSSION] okay, mp3 fans...

[Jonathan King <EMAIL:PROTECTED>]

On Wed, 31 Oct 2001, Mike Miller wrote:

> On Wed, 31 Oct 2001, Igor Izyumin Jr. wrote:
>
> > Mike Miller wrote:
> > > But no one would be able to tell (by ear).  Right?
> >
> > I am sure that an expert could distinguish a CD and a more high-
> > definition format.
>
> How do you know?  Has anyone proved this in a controlled study?

OK, so my opportunity to do a lit search on this right now is limited.
What I do know (or have been reminded of) are the following tidbits.

1) People can use more than 16 bits of dynamic range in their sound input.
   Now this would only be a curiosity except that there are pieces of
   music that could require this in terms of overall sound levels
   (especially classical pieces).  It is also the case that musical notes
   and human voices both contain a wide dynamic range in their upper
   harmonics, and you can certainly construct stimuli that are
   discriminable on that basis.  Note that if we had 20 or (certainly) 24
   bits of information in the standard, these concerns would both be moot;
   we'd be past the ability of humans to tell the difference, and probably
   past the point where there was any utility even to, e.g., a
   computer-based system trained to respond only to your voice (for
   example).  I bring up the latter point because it is not at all
   unreasonable to expect machines to be able to use a different set of
   cues to perform tasks like voice identity recognition.  And then
   there's arguments for more bits that depend on steganography, indexing,
   or close-captioning.

2) The frequency range of human hearing is highly variable between
   individuals and within individuals across their life-span.  So Mike
   Miller's hearing used to be great, then he started to blast "Whole
   Lotta Love" into stereo headphones, which took its toll.  And then
   he got old, which not only meant he had to turn the volume on his
   'Zepplin up to 11 to get the same effect, but ramp up the treble as
   well.  In any case, the upper frequency limits for most of us are in
   the 15-20 kHz range *for pure sine waves*.  Moreover, you need a lot
   of power at the upper end of the range to get the same effect as at
   (say) 3 kHz.  We all probably know that you need to sample at at least
   double the highest frequency you want to recreate, and some of us may
   be aware that a bit of oversampling never hurts in actual
   reconstruction.  At 44 kHz, the CD standard is uncomfortably close to
   the minimum specs, you'd like to hit, and this is especially true
   when you consider point 3.

3) Real sounds have content that exceeds the range of human hearing, and
   that means you have to be careful when you record them, since if you
   *don't* or *can't* 2x sample the highest frequencies, they just get
   aliased into lower frequency bands.  (You can prove this to yourself
   with a paper and pencil.  Just draw a sine wave, and then draw ticks
   on it that are more widely spaced than half the period of the wave;
   connect the ticks, and note that you've sampled a sine wave with a
   much lower frequency than the one you've started with.)  This could
   have been a problem with very naive mastering of early CDs: you take
   your original, sample it down to 44 kHz without low-pass-filtering it
   first, and you lose.  The fix there is easy, of course.  What is more
   important, however, is that even the best low-pass filters have a
   significant degree of "roll-off", so to get a signal that's "full
   power" from 20 Hz to (say) 16 kHz with a modest amount of oversampling
   could require more bandwidth than the 44 kHz you get with CDs.

4) Another problem with CDs versus vinyl that's for real is the fact that
   it's much tougher for a needle to reproduce the highest frequencies on
   a vinyl LP for physical reasons, so that if you're mastering
   something for vinyl, you want to systematically boost the treble to
   compensate for this.  Now, if you take the same master and naively
   drop it onto a CD, it *WILL* sound tinny.  Because it *is* tinny
   because you were mastering for vinyl and its particular frequency
   response curve.  Now, this gets even more insidious, since some of the
   treble boost probably happened at the time of the original
   (analog) recording, and now you've got an interesting problem.

   And then there's the guy who tunes his system to perfect the replay
   of his vinyl recordings, and then tries to play a CD...with less than
   stellar results.

> Many people believe things that are totally false (e.g., see the green
> pen stuff on the net).  Back when CDs were new, many people claimed
> that their sound quality was not as good as vinyl, but I've also heard
> that those people could not tell the difference in a properly
> controlled study.

Making this more difficult is the fact that the earliest CDs probably were
deficient, because they were stomped from masters tuned for vinyl, while
other (classical) CDs will always be dependent on weirdly tuned analog
masters that will sound different on CDs.  But you're right: the
overwhelming majority of people don't (and can't) care.  Except maybe for
the dynamic range thing...I think that is probably a real problem.

> What you are saying makes some sense, but I won't be convinced until I
> read of some proper research on the issue.  Another thing to bear in
> mind (as someone else, maybe Jon, mentioned) is that recording
> engineers may need to store a better representation of the sound for
> editing purposes so that the final product will have an appropriate
> effective bitrate.

Also note that *you* want that higher bitrate, too, if you're going to do
something like rip the tracks off the source and mp3-ify them.  My guess
is that some of the problems with lower bit rate mp3s would actually go
away if the encoders could see more information up front.

> > This is the same argument as regular TV vs. HDTV
>
> Not at all -- *anyone* can *easily* see the difference in the TVs, but
> it takes a really good ear to tell the difference in the digitally
> encoded sounds you are talking about (if it can be done at all).

Do note, however, that a large source of the improvement between TV and
HDTV is not dependent on the superior resolution, but the fact that NTSC
itself is a really sucky transmission standard for color information where
literally everything that can go wrong, does go wrong over the air or
through cable.  I haven't heard of many comparisons between a standard set
playing a DVD and the same result on a (comparably featured!) HDTV set; I
expect that the difference would be much less noticable.

jking

--
To unsubscribe, go to http://mlug.missouri.edu/members/edit.php

Archives are available at http://mlug.missouri.edu/list-archives/