Newsgroups: comp.speech
Path: pavo.csi.cam.ac.uk!doc.ic.ac.uk!decwrl!decwrl!uunet!munnari.oz.au!bruce.cs.monash.edu.au!trlluna!titan!titan!duke
From: duke@titan.trl.OZ.AU (Peter Duke)
Subject: Re: mu law to linear conversions
Message-ID: <1993Aug6.070616.23721@trl.oz.au>
Sender: root@trl.oz.au (System PRIVILEGED Account)
Organization: Telecom Research Labs, Melbourne, Australia
References: <1993Aug3.144124.20966@afterlife.ncsc.mil>
Date: Fri, 6 Aug 1993 07:06:16 GMT
Lines: 38

jpcampb@afterlife.ncsc.mil (Joe Campbell) writes:
> In article <1993Aug3.130944.29274@gps.leeds.ac.uk>,
> S C Arnfield <csc5sca@gps.leeds.ac.uk> wrote:
> >In article <23kkc3$8cv@vtserf.cc.vt.edu>, bhushan@mprg1.mprg.ee.vt.edu (Bhushan Rele) writes:
> >> Hi
> >> I have been using the SUN workstations audio facility to record some
> >> stuff and convert it into linear PCM using MATLAB for further processing
> >> by a vocoder.
> >> 
> >> The vocoder requires minimum 13 bit quantization.
> >> 
> >> Since the *.au files of the SUN are 8 bit mulaw, will the conversion
> >> result in a loss of precision?Should I try something else for my work
> >> in the future? 

Alaw and mulaw are defined in CCITT (now Telecom. Standard. Sector) Rec.
G.711. To implement them strictly, you would need 13 and 14 bits in the
A/D converter respectively, although you could get by with one less bit
if you offset the A/D by 1/2 LSB so as to be symmetrical about zero 
(i.e. no true zero).

Naturally if data is represented by 8 bits and no attempt is made to
exploit redundancy (as is the case with Alaw and mulaw), there will 
be loss of precision.  Alaw and mulaw are piecewise logarithmic coders, 
which means that for small amplitudes there is no loss, while for large 
amplitudes the ABSOLUTE quantizing distortion is allowed to increase 
such that the RELATIVE quantization distortion (ie signal to noise ratio) 
is kept constant.

If you are only concerned with good telephone quality then rest
assured that 8 bit mulaw will suffice. If you are concerned with HiFi
music then that is another matter.  If the vocoder is for use in the
telephone network then chances are that the speech will have gone
through such a coding process anyway.

Peter Duke
p.duke@trl.oz.au

