ipod recording

[Natasha L Warner]

Hi,

before we totally write off all compressed data that already exists, let's
think twice.  Sure, if you're choosing at the beginning of a project, it's
probably far better to just record uncompressed in the first place.  But
if data exists that was recorded in a compressed format, that doesn't make
it useless, it depends greatly on what you're going to be measuring.
Pitch, for example, is extremely robust, even under atrocious recording
conditions and with poor equipment.  One of the few things that will make
pitch unmeasurable is overlapping voices (or other periodic sound sources)
where the one you're trying to measure is not substantially louder than
the others.  But considering that you can measure pitch even with shocking
amounts of background noise, or an extremely clipped signal, I'd be
surprised if you couldn't measure it pretty reliably on mp3 data.  On the
other hand, if you want to measure voice quality (creaky voice, breathy
voice, etc.), then you need to know that you have a really reliable
recording that didn't mess with the spectral tilt.  But at that point,
position of the mic relative to the mouth is also going to be pretty
important.

A few years ago when the compression at issue was minidiscs, I
heard that a few phoneticians, including Ian Maddieson, did a test of
compressed vs. uncompressed recordings and couldn't find a difference in
any part of the signal they were interested in.  They concluded the
compression wasn't so bad.  Of course, I'd rather know I have uncompressed
data so I just don't have to worry about it.  But given a choice of using
existing compressed data or starting over, I would look at the signal and
see what looked measurable and what didn't, keeping the compression in
mind.

Also, it's worth considering that in many field recording situations, the
background noise (whether it's airplanes, construction noise, chickens,
kids, crickets, refrigerator hum, a college cafeteria, or a preschool
setting--I've run into all of these at least indirectly, for acoustic
phonetics work) is going to contribute a whole lot more problems to the
signal than the compression creates.  And sometimes we do work with data
collected over the phone, which limits the signal a lot more than
compression does.  So my personal approach would be to use a reasonably
good headmounted mic, get it as close to the speaker's mouth as possible,
make sure any other noise sources are far further from the mic than the
speaker's mouth is, and record uncompressed if possible.  It would also be
nice to have a clearer understanding than I do at this point of just what
we lose during the compression.

Thanks for the discussion, everyone,

Natasha Warner

*******************************************************************************
Natasha Warner
Associate Professor, Department of Linguistics
University of Arizona
PO Box 210028
Tucson, AZ 85721-0028