Path: utzoo!utgpu!news-server.csri.toronto.edu!rpi!zaphod.mps.ohio-state.edu!uwm.edu!bionet!agate!ziploc!eps
From: eps@toaster.SFSU.EDU (Eric P. Scott)
Newsgroups: comp.sys.next
Subject: Re: Soundfiles
Message-ID: <1451@toaster.SFSU.EDU>
Date: 31 Mar 91 04:54:11 GMT
References: <17776@sdcc6.ucsd.edu> <1991Mar26.004133.4930@unx2.ucc.okstate.edu>
Reply-To: eps@cs.SFSU.EDU (Eric P. Scott)
Organization: San Francisco State University
Lines: 42

In article <1991Mar26.004133.4930@unx2.ucc.okstate.edu>
	minich@unx2.ucc.okstate.edu (Robert Minich) writes:
>How about a reference that explains mulaw encoding for those of use
>without these info sources.

Mu-law encoding is a form of nonuniform companding used for
digitized voice in U.S. telephony.  8 bit mu-law provides a
small-signal S/N ratio and dynamic range roughly equivalent to a
12-bit linear representation--thus it can be thought of as a data
compression technique.  [In practice, it *is* losing information
(and the difference between the original and the encoded version
is digital noise); it's designed to get 50% more "value" out of a
communications channel without sacrificing too much speech
intelligibility.  It's not something you'd want to use for
arbitrary audio.]  One property is does share with compressed
data is that "you can't do anything useful with it" without
expanding it first.

mu-law converts normalized real amplitudes according to

		input: 0 <= |x| <= 1

		    log ( 1 + {mu} |x| )
		y = --------------------
		      log ( 1 + {mu} )

		parameter {mu} = 255

[8-bit mu-law is a fixed-point representation, and something I
read suggests that it may actually be a 15-segment linear
approximation to the above.  I confess to not quite doing my
homework on this one; I need to track down a copy of CCITT
Recommendation G.711 next.]

Anyway, since mu-law quantization levels are more closely spaced
at low amplitudes, their S/N ratio improves, while higher
amplitudes tend to mask quantization noise.  Low frequencies with
changing amplitudes would have the most perceptible problems, but
the nominal bandpass filtering used with 8KHz sampling attenuates
frequencies below 300Hz ... so we stay out of trouble.

					-=EPS=-