This article is from the CD-Recordable FAQ, by Andy McFadden (firstname.lastname@example.org) with numerous contributions by others.
The "Red Book" specification for audio CDs chose 44100 samples per second,
where each sample is 16-bit stereo PCM. PCM is a fine choice for encoding
audio, stereo is widely recognized and supported, and it's very easy to
manipulate data in 16-bit quantities with existing hardware and software.
Why 44100? Why not make it a round decimal value like 44000, or a round
binary quantity like 44032? Why not 32KHz or 48KHz?
In general, the human ear can hear tones out to about 20KHz. According
to a smart fellow named Nyquist, you have to sample at twice that rate.
Because of imperfections in filtering, you actually want to be a little
According to John Watkinson's _The Art of Digital Audio_, 2nd edition, page
104, the choice of frequency is an artifact of the equipment used during
early digital audio research. Storing digital audio on a hard drive was
impractical, because the capacity needed for significant amounts of 1 Mbps
audio was expensive. Instead, they used video recorders, storing samples
as black and white levels. If you take the number of 16-bit stereo samples
you can get on a line, and multiply it by the number of recorded lines in
a field and the number of fields per second, you get the sampling rate.
It turned out that both NTSC and PAL formats (the video standards used in
US/Japan and Europe, respectively) could handle a rate of 44100 samples per
second. This rate was carried over into the definition of the compact disc.
The sampling rate for "professional" audio, 48KHz, was chosen because it's
an easy multiple of frequencies used for other common formats, e.g. 8KHz
for telephones. It also happens to be fairly difficult to do a good
conversion from 48KHz to 44.1KHz, which makes it harder to, say, copy an
audio CD with a "consumer" DAT deck. (Well, okay, some consumer DAT
decks can do 44.1KHz now, but initially only "professional" decks could
handle the lower frequency.)
There is relatively little difference in audible quality between 44.1KHz
and 48KHz, since the slight increase in frequency response is outside the
range of human hearing. Some inaudible tones produce "beats" with audible
tones and thus have a noticeable impact, but the improvement from 44.1 to
48 is marginal at best.