🚀 go-pugleaf

We're developing a voice annotation application and would like to
stick with just one of the many compression algorithms.  What's *your*
favorite and why?  (I'll compose an article based upon your responses.)

Goal: We'll be sampling at 8kHz and need to reduce the data to 4 bits
per sample.

Possibilities: The "contestants" appear to be "scaled MULAW" and ADPCM.

"Scaled MULAW" is typified by the Artisoft "Sounding Board" hardware
and software.  They start with 13-bit PCM, compand to 8-bit MULAW
(which has the advantage of being in the public domain), then compress
approximately 2:1 again by converting every 32 samples to 4 bits each
plus a "scaling" byte.  (A proprietary but simple algorithm.)

"ADPCM" is typified by Sound Blaster (Creative Labs) and others that
start with either 8- or 12-bit samples then reduce to 4-bit samples
using ADPCM.  The primary disadvantage of this technique (it seems to
us) is the lack of standards for ADPCM.
 
Given that background, which technique would *you* use?  Do you have an
opinion as to which "sounds better", Scaled MULAW or ADPCM?  What are
your personal experiences either as a user or developer/engineer?
 
Thanks!      ...doug


==================================================
Doug Kaye   Rational Data Systems   San Rafael, CA
doug@rds.com   {moon|uunet}!rds!doug  MHS:doug@rds
==================================================