[codec] Transparency or intelligibility?

[Gregory Maxwell <[e-mail]>]

One source of disagreement over requirements which I've seen crop up with respect to audio
codecs arises out of what appears to be a lack of agreement on the end goal: Is the expectation 
that the result will be transparent, or that it will be merely intelligible and/or 'good sounding'?

By transparent I mean something roughly like "an experienced listener is unable to distinguish in a
blind listening test even with material selected to be especially challenging for the codec".

While it's is not possible to deliver transparency without intelligibility, it is possible to deliver excellent
intelligibility without a hope of transparency.   So optimizing for one vs the other can lead to different
engineering trade-offs.

For example,  a 7kHz low-pass will produce speech with perfect intelligibility but at the same time
would have absolutely no hope of being transparent.  If you had a goal of even near transparency
you would probably not even consider a codec with a low-pass lower than about 18kHz or so.
(For general audio— for pure speech and adult listeners you could possibly go a bit lower but 7kHz is
far too low)

It was my thought that for the transmission of sound typical Internet bandwidths are high enough that
we really have little excuse but to provide transparency under good operating conditions.

Of course, during levels of high loss or congestion transparency will not be possible and in that case
a codec should maintain intelligibility under adverse conditions and that should be an absolute
requirement too, but in the normal case transparency should be expected.

Is this thinking consistent with the thoughts of other working group participants?