Date: Wed Oct  1 10:07:09 CDT 1997


I have an enhanced version of the U.S. D.O.D. LPC-10 speech coder
software available for use.  You can get it here:

http://www.arl.wustl.edu/~jaf/lpc/lpc10-1.5.tar.gz

This lossy speech coder compresses 8 KHz sampled audio to 2400 bits
per second with what I'd call reasonable quality of reconstructed
speech.  It takes about 50% of my 486DX2-66 CPU to compress in real
time, and about 25% of the same CPU to decompress in real time.
Sparc-2 speed tests give similar results.  I've used it with good
results in the encrypted Internet phone application Nautilus (see
http://www.lila.com/nautilus/).

The last time I announced it was what I called release 1.4.  Not much
has changed since then, but here is a summary of everything that has
changed since release 1.2.  For those who might want to use it in an
Internet MBONE audio application, be sure to read the 1.4 changes.

Release 1.3: Added ready-to-compile C sources (at least for Unix
systems with GNU make and gcc) so you don't have to figure out how to
install the Fortran to C conversion program f2c.

Release 1.4: This release includes hand-modified C sources that allow
an application writer to compress (or decompress) multiple audio
streams "in parallel", by interleaving calls to the encoding (or
decoding) functions for each frame (22.5 ms), and keeping a separate C
struct for each audio stream for state that must be maintained from
one frame to the next.  This makes it much handier for use in Internet
MBONE audio tools that can simultaneously receive audio packets from
more than one source.

Release 1.5: Fixed a couple of minor bugs in the nuke and unnuke
sample applications that would cause problems if your compiler used 16
bit integers for C "int" variables.  This was only a bug in the sample
application code, not in the LPC-10 library itself.

Andy Fingerhut
jaf@arl.wustl.edu
Washington University, St. Louis MO
WWW home page: http://www.arl.wustl.edu/~jaf/